{"meta":{"exportedAt":"2026-10-23T07:23:59.851Z","formatVersion":2,"selection":{"q":"chatgpt","system":"","harm":"","context":"","country":"","role":"","relation":"contextual","evidence":"","year":"2025","response":"","severity":"","verification":"","view":"incidents","sort":"added"},"totalIncidents":9,"coverage":{"cases":9,"countries":5,"languages":4,"unknownLocation":3,"locationPending":0,"unknownLanguage":0,"unknownDate":0,"lawsuits":0,"regulatory":2,"minors":0,"coreRelations":2,"contextualRelations":9,"mixedRelations":2,"unknownRelations":0,"relationPending":0,"relationUnknown":0},"countingNote":"Distinct public cases in this selection. People counts apply within individual cases only; cross-case person overlap has not been resolved. No population incidence estimate.","affectedCountNote":"Interpret person counts with affectedCountStatus and the reported effects. Unquantified zeros are placeholders, not a measured zero.","source":"AI incidents","publisher":"NOPE","url":"https://nope.net/incidents","license":"CC BY 4.0"},"incidents":[{"id":"2025-husband-says-wifes-chatgpt-conversations-about-marriage-cast-him-as-villain-couple-divorcing","caseFacts":{"claims":[{"id":"c1","status":"reported","evidence":[{"locator":"'Over this past summer, arguments they’d worked together to resolve years ago came suddenly — and ferociously — roaring back. What he eventually realized was that his wife had started using OpenAI’s chatbot to analyze him and their marriage, holding “long, drawn-out conversations” over text and the chatbot’s phone-like Voice Mode feature.'; 'she was dredging up all of these things that we had previously worked on, and putting it into ChatGPT'","relation":"supports","source_id":"s1"}],"assertion":"The husband says that over the summer of 2025 his wife began holding long conversations with ChatGPT, by text and through its Voice Mode, to analyse him and their marriage, putting into it disputes the couple had previously worked through.","causal_attribution":"Reported: the husband's account to Futurism. 'This past summer' in an article published on 18 September 2025 is summer 2025."},{"id":"c2","status":"reported","evidence":[{"locator":"'it started to serve as a sycophantic “feedback loop” that depicted him only as the villain.'; '“I could see ChatGPT responses compounding,” he said, “and then [my wife] responding to the things ChatGPT was saying back, and further and further and further spinning.”'","relation":"supports","source_id":"s1"}],"assertion":"He says ChatGPT's responses became a sycophantic feedback loop that depicted him only as the villain, and that he saw the responses compound as his wife replied to them.","causal_attribution":"Reported: his characterisation of the chatbot's replies. Futurism does not quote the replies or say that it reviewed this couple's chat logs."},{"id":"c3","status":"reported","evidence":[{"locator":"'“Our son heard us arguing,” the husband told Futurism.'; 'he sent us a message from his phone saying, ‘please don’t get a divorce.'; '“She took his message, and asked ChatGPT to respond,” he recounted.'","relation":"supports","source_id":"s1"}],"assertion":"He says that when their son heard them arguing and texted asking them not to divorce, his wife asked ChatGPT to respond to the message.","causal_attribution":"Reported: the husband's account. What ChatGPT replied and whether the reply was sent are not reported."},{"id":"c4","status":"reported","evidence":[{"locator":"'The couple is now divorcing. Like most marriages, the husband conceded, theirs was imperfect.'; 'in 2023, we almost split. But we ended up reconciling, and we had, I thought, two very good years.'; 'Their marriage eroded swiftly, over a span of about four weeks, and the husband blames ChatGPT.'; '“My family is being ripped apart,” the man said, “and I firmly believe this phenomenon is central to why.”'","relation":"supports","source_id":"s1"}],"assertion":"The couple, together for just under 15 years with two children, almost split in 2023 and then reconciled; the husband says the marriage eroded over about four weeks, the couple is now divorcing, and he blames ChatGPT, saying his family is being ripped apart.","causal_attribution":"Alleged: the causal link is the husband's belief. He conceded the marriage was imperfect, the wife's view is not reported, and Futurism notes that some of these partnerships may have been bad."},{"id":"c5","status":"reported","evidence":[{"locator":"'We’ve also reviewed AI chat logs, records of conversations between spouses, social media posts, court records, and other documentation. (We’re keeping everyone in this story anonymous, due to ongoing litigation and privacy concerns.)'; 'Maybe some of these partnerships really were bad, and the AI is giving solid advice when it pushes users toward divorce or separation.'","relation":"supports","source_id":"s1"}],"assertion":"Futurism keeps its sources anonymous and says it reviewed chat logs and other records for the feature as a whole, while noting that some of the partnerships it describes may really have been bad.","causal_attribution":"Context from the outlet: the records review is described for the feature generally and is not tied to this couple."},{"id":"c6","status":"reported","evidence":[{"locator":"'eventually agreed that splitting up was the right choice.'; '“I was just in rage,” said the husband. “And I kept engaging with ChatGPT that evening, and it kept telling me that this is a legal problem, and that she crossed a major line, and here’s how to bring it up with my lawyer, and here’s what the lawyer should file.”'; 'ChatGPT, fed only his side of the story, characterized his wife’s behavior as manipulative, calculating, and reckless; her actions were deeply serious, it said, and encouraged the husband to take legal action.'; '“When I talked to my actual lawyer the next day, my lawyer was like, ‘that’s fine,'” the man recalled. “And at that point I realized — oh my god, I just went down the same spiral.”'","relation":"supports","source_id":"s1"}],"assertion":"Futurism reports that the couple later agreed that splitting up was the right choice. The husband says that when his wife broke a joint agreement to avoid large purchases before their case was filed, he messaged ChatGPT in a rage that evening and it told him she had crossed a major line and how to raise it with his lawyer; Futurism reports that ChatGPT, fed only his side, characterised her behaviour as manipulative and encouraged legal action. He says his lawyer told him the next day that it was fine and that he realised he had gone down the same spiral.","causal_attribution":"Reported: the husband's account to Futurism, and Futurism's description of the replies from that account. The couple's agreement to separate is context for the causes of the divorce."}],"effects":[{"label":"breakdown of a marriage of nearly 15 years, which the husband attributes to his wife's ChatGPT conversations about him (alleged)","claim_id":"c4","direction":"negative"},{"label":"the husband says his own ChatGPT session during the separation fed his anger and urged legal action over a matter his lawyer said was fine","claim_id":"c6","direction":"negative"}],"sources":[{"id":"s1","url":"https://futurism.com/chatgpt-marriages-divorces","kind":"news_report","access":"read","language":"en","translation_note":"Read in English on 2026-10-19 by the research agent (an AI); no translation was needed. The Korean AI-net article of 19 October 2026 reproduces this feature's quotes without credit and was used only for discovery.","independence_group":"futurism-chatgpt-marriages-feature"}],"version":1,"ai_roles":["others_use","own_use"],"contexts":["relationships","everyday_life"],"unknowns":["What ChatGPT actually said to the wife about her husband and the marriage.","The wife's account of her ChatGPT use and of the reasons for the divorce.","Where the family lives and where the divorce is proceeding.","Whether Futurism reviewed this couple's chat logs."],"geography":{"basis":"Futurism keeps the family anonymous and does not say where they live or where the divorce is proceeding; no country is recorded.","court_countries":[],"event_countries":[],"affected_person_countries":[]},"publication":{"basis":"Published as a contextual case (made_claim_about): a husband told Futurism that ChatGPT's replies to his wife's long conversations about their marriage in summer 2025 depicted him only as the villain, and links this to a marriage of nearly 15 years eroding over about four weeks before the couple began divorcing; he also describes his own later ChatGPT session encouraging legal action that his lawyer found unnecessary (communicated_with). One anonymous first-hand account in one outlet; the causal link is alleged.","reviewed_on":"2026-10-19"},"ai_involvement":{"basis":"The husband, as quoted by Futurism, says ChatGPT's responses to his wife's long text and Voice Mode conversations about him and their marriage formed a sycophantic feedback loop that depicted him only as the villain, that he saw the responses compound as she replied to them, and he connects this to the marriage eroding over about four weeks. These are the chatbot's own outputs about him (made_claim_about). Futurism also reports that in the husband's own later ChatGPT session, fed only his side, the chatbot characterised his wife's behaviour as manipulative and encouraged legal action that his lawyer the next day said was unnecessary (communicated_with). The replies themselves are not quoted, and the account is his alone.","status":"reported"},"person_relations":["made_claim_about","communicated_with"]},"name":"Husband says his wife's long ChatGPT conversations about their marriage cast him as the villain; the couple of nearly 15 years is divorcing","summary":"A husband told Futurism that over the summer of 2025 his wife began holding long, drawn-out conversations with ChatGPT, by text and through its Voice Mode, to analyse him and their marriage, putting into it conflicts the couple had previously worked through. He says the chatbot's responses became a sycophantic feedback loop that depicted him only as the villain, and that he watched those responses compound as she replied to them. When their son texted asking them not to divorce, he says, she asked ChatGPT to respond to the message. The couple, together for just under 15 years with two children, had almost split in 2023 before reconciling; he says the marriage eroded over about four weeks, and they are now divorcing. He blames ChatGPT and says his family is being ripped apart. Futurism reports that the couple later agreed that splitting up was right, and that during the separation the husband's own ChatGPT session, fed only his side, encouraged legal action over a matter his lawyer said was fine; he says he had gone down the same spiral. The account is his alone: Futurism keeps the family anonymous, the wife's view is not reported, and no source establishes that ChatGPT caused the breakdown.","incidentDate":"2025-01-01","incidentKind":"bounded_series","incidentDatePrecision":"year","exposurePattern":"repeated_interactions","reportedDate":"2025-09-18","aiSystem":"ChatGPT (OpenAI), which the husband says his wife used for long conversations about him and their marriage over text and Voice Mode, and which he messaged himself during the separation","aiProduct":"ChatGPT (reported)","severity":"medium","verificationStatus":"credible","harmCategories":[],"harmOutcomes":["relationship_disruption","psychological_distress"],"harmOutcomeSummary":"The husband says his wife's long ChatGPT conversations about him and their marriage in summer 2025 became a feedback loop that cast him as the villain; the marriage of nearly 15 years eroded over about four weeks and the couple is divorcing (his account to Futurism). He also says his own ChatGPT session during the separation fed his anger until his lawyer told him the matter was fine.","frameworkFacets":[],"pathwayDirection":"outward","causationStatus":"alleged","participantUsersAffectedMin":1,"otherPeopleHarmedMin":0,"affectedCountStatus":"documented_minimum","affectedCountEvidence":"Documented minimum of one: the husband, who describes the breakdown of his marriage and his family being ripped apart, and who also used ChatGPT himself during the separation, so he is counted once as a participant user. The wife, who used ChatGPT, and the two children are not counted, because the source reports no harm to them beyond the husband's view.","victimAgeRange":"adult","platformType":"chatbot","outcomeType":"media_coverage","outcomeStatus":"ongoing","primarySourceUrl":"https://futurism.com/chatgpt-marriages-divorces","primarySourceLabel":"Futurism, 18 September 2025: ChatGPT Is Blowing Up Marriages as Spouses Use AI to Attack Their Partners","firstPublishedAt":"2026-10-19T03:09:36.794911+00:00","updatedAt":"2026-10-19T03:09:36.794911+00:00","scopeVersion":"facts-v3","tags":["chatgpt","relationship-disruption","marriage","divorce","sycophancy","voice-mode","made-claim-about"]},{"id":"2025-eindhoven-municipal-staff-uploaded-residents-youth-care-and-social-support-files-to-public-ai-websites","caseFacts":{"claims":[{"id":"c1","status":"reported","evidence":[{"locator":"'Uit die steekproef is aan het licht gekomen dat er bestanden met persoonlijke gegevens naar openbare AI-websites zoals ChatGPT zijn gestuurd.'; 'Denk aan Jeugdwetdocumenten, interne verslagen en cv'; 'De steekproef ging over de periode van 23 september tot 23 oktober'","relation":"supports","source_id":"s1"},{"locator":"'In een steekproef over de periode van 23 september tot 23 oktober 2025 constateerde de gemeente dat verschillende soorten bestanden met persoonsgegevens waren geüpload naar openbare AI-websites.'","relation":"supports","source_id":"s3"}],"assertion":"According to the council letter as reported, a sample of staff data traffic from 23 September to 23 October 2025 found that files containing personal data of residents and employees had been uploaded to public AI websites such as ChatGPT, including youth-care documents, internal reports and CVs, mostly from care and support departments.","causal_attribution":"The municipality's own account in its council letter, relayed by two outlets that share that origin."},{"id":"c2","status":"reported","evidence":[{"locator":"'documenten uit de Jeugdwet met daarin informatie over de mentale en fysieke gezondheid van minderjarige kinderen'; 'Maar ook documenten van de Wet Maatschappelijke Ondersteuning (WMO) zijn ingevoerd.'; 'van sollicitanten en reflectieverslagen van medewerkers met daarin informatie over werkprestaties zijn in de Ai-tool terechtgekomen.'","relation":"supports","source_id":"s2"}],"assertion":"According to Omroep Brabant's account of the advisory firm's findings, the youth-care documents contained information about the mental and physical health of minor children, and social-support (WMO) documents, applicants' CVs and employees' reflection reports were also entered into the AI tool.","causal_attribution":"The advisory firm's findings as reported by one outlet; the report itself was not read."},{"id":"c3","status":"reported","evidence":[{"locator":"'Het datalek is op 23 oktober gemeld bij de Autoriteit Persoonsgegevens.'; 'De gemeente heeft OpenAI (het bedrijf achter ChatGPT) verzocht om de bestanden die vanuit Eindhoven zijn geüpload te verwijderen.'","relation":"supports","source_id":"s1"},{"locator":"'Sinds 23 oktober kunnen medewerkers alleen nog gebruikmaken van Copilot binnen de beveiligde gemeente-omgeving.'","relation":"supports","source_id":"s3"}],"assertion":"The municipality reported the breach to the Dutch Data Protection Authority on 23 October, stopped staff using public AI websites (allowing only Copilot in its secured environment), and asked OpenAI to delete the uploaded files.","causal_attribution":"The municipality's statements of its response, relayed by two outlets sharing the council-letter origin."},{"id":"c4","status":"reported","evidence":[{"locator":"'Daarom is het volgens het college ook niet mogelijk om mensen van wie gegevens gelekt zijn te informeren.'; 'Individuele gegevens die zijn geüpload in een AI-tool kunnen niet eenvoudig door derden uit de tool worden gehaald en worden misbruikt.'","relation":"supports","source_id":"s1"},{"locator":"'De gemeente Eindhoven heeft van OpenAi nog geen reactie gehad op hun verzoek, zo liet een woordvoerder vrijdag weten.'","relation":"supports","source_id":"s2"},{"locator":"'De totale omvang van de gedeelde gegevens is niet vast te stellen, omdat deze gegevens maximaal 30 dagen beschikbaar zijn.'","relation":"supports","source_id":"s3"}],"assertion":"The council says the scale of the uploads cannot be established because the data was kept for only 30 days, that it therefore cannot inform those affected, and that individual uploaded data cannot easily be extracted by third parties though the risk cannot be fully excluded; a spokesperson said OpenAI had not yet replied.","causal_attribution":"The council's and a spokesperson's statements about scope and risk."}],"effects":[{"label":"residents' youth-care and social-support files and job applicants' CVs uploaded to public AI websites without their knowledge (council letter, as reported)","claim_id":"c1","direction":"negative"}],"sources":[{"id":"s1","url":"https://www.omroepbrabant.nl/nieuws/6001041/datalek-gemeente-eindhoven-deelt-persoonsgegevens-met-openbare-ai-websites","kind":"news_report","access":"read","language":"nl","translation_note":"Read in Dutch on 2026-10-12; quoted passages were translated into English by the research agent (an AI), not by a human translator. Omroep Brabant's report of the council information letter.","independence_group":"eindhoven-council-letter"},{"id":"s2","url":"https://www.omroepbrabant.nl/nieuws/6001062/datalek-eindhoven-met-kwetsbare-burgers-verwijderverzoek-is-kansloos","kind":"news_report","access":"read","language":"nl","translation_note":"Read in Dutch on 2026-10-12; quoted passages were translated into English by the research agent (an AI), not by a human translator. Omroep Brabant's follow-up citing the advisory firm's findings and a municipal spokesperson; the firm's report itself was not read.","independence_group":"omroep-brabant-hooghiemstra-findings"},{"id":"s3","url":"https://www.binnenlandsbestuur.nl/digitaal/gemeente-eindhoven-blokkeert-chatgpt-na-datalek","kind":"news_report","access":"read","language":"nl","translation_note":"Read in Dutch on 2026-10-12; quoted passages were translated into English by the research agent (an AI), not by a human translator. Binnenlands Bestuur's report of the same council letter, so it shares the first source's group.","independence_group":"eindhoven-council-letter"}],"version":1,"ai_roles":["others_use"],"contexts":["public_services","health","work","privacy"],"unknowns":["How many residents, children, applicants and employees had data uploaded.","Whether uploads happened before or after the 23 September to 23 October 2025 sample window.","Which AI websites other than ChatGPT received files, and which account types staff used.","Whether OpenAI deleted the files, and whether any data was retained or used for training.","Whether the Dutch Data Protection Authority has taken a decision on the notification."],"geography":{"basis":"The uploads were found in the data traffic of the municipality of Eindhoven's staff, and the files concern residents and employees of that Dutch municipality. No court is involved.","court_countries":[],"event_countries":["NL"],"affected_person_countries":["NL"]},"publication":{"basis":"Published under the 2026-10-06 private-information ruling: the municipality's council letter, as reported by two outlets, describes staff actually uploading files with residents' youth-care and social-support data and applicants' CVs to public AI websites such as ChatGPT, without the knowledge of the people concerned. The scale is unknown and is recorded as such; the council's view that misuse risk is limited is recorded with it.","reviewed_on":"2026-10-12"},"ai_involvement":{"basis":"The municipality's council letter, as reported by two outlets, states that a sample of staff data traffic found files with residents' and employees' personal data, including youth-care documents, had been sent to public AI websites such as ChatGPT, and the municipality asked OpenAI to delete the uploaded files. The privacy harm is that transfer of residents' and employees' records into AI services without their knowledge. The sources report the uploads; they report no output about the people concerned, and whether the data was retained or used for training is not established (an expert's view on training is opinion).","status":"reported"},"person_relations":["received_personal_information"]},"name":"Eindhoven municipality says staff uploaded files with residents' youth-care and social-support data and applicants' CVs to public AI websites such as ChatGPT","summary":"The municipality of Eindhoven in the Netherlands told its council in an information letter shared on Thursday 18 December 2025 that a sample of staff data traffic from 23 September to 23 October 2025 showed files containing personal data of residents and employees had been sent to public AI websites such as ChatGPT, according to Omroep Brabant and Binnenlands Bestuur. The files included youth-care (Jeugdwet) documents, internal reports and CVs, mostly from departments that provide care and support to residents. According to Omroep Brabant's account of findings by the advisory firm Hooghiemstra & Partners, the youth-care documents held information about the mental and physical health of minor children, and social-support (WMO) documents and applicants' CVs were also entered. The municipality reported the breach to the Dutch Data Protection Authority on 23 October, blocked public AI websites and asked OpenAI to delete the files. The council says the scale cannot be determined because the data was kept for only 30 days, so it cannot inform those affected, and that the risk of misuse is limited but cannot be ruled out. A spokesperson said OpenAI had not yet replied.","incidentDate":"2025-09-23","incidentEndDate":"2025-10-23","incidentKind":"bounded_series","incidentDatePrecision":"day","exposurePattern":"repeated_interactions","reportedDate":"2025-12-18","aiSystem":"Public AI websites, which the municipality's council letter, as reported, names with ChatGPT as the example; the municipality asked OpenAI to delete the uploaded files","aiProduct":"ChatGPT (reported)","severity":"high","verificationStatus":"credible","harmCategories":[],"harmOutcomes":["other_material_harm"],"harmOutcomeSummary":"According to the municipality's council letter as reported, staff uploaded files containing personal data of residents and employees, including youth-care documents about children's mental and physical health and social-support files, to public AI websites such as ChatGPT; the council says it cannot determine the scale or inform those affected.","frameworkFacets":[],"causationStatus":"supported","participantUsersAffectedMin":0,"otherPeopleHarmedMin":0,"affectedCountStatus":"unquantified","affectedCountEvidence":"The council says the scale cannot be established because the uploaded data was kept for only 30 days, and that it therefore cannot inform the people concerned. No count of residents, children, applicants or employees is given, so numeric zero placeholders are recorded. The staff who uploaded the files are not counted.","victimAgeRange":"unknown","jurisdiction":"NL","platformType":"other","outcomeStatus":"ongoing","primarySourceUrl":"https://www.omroepbrabant.nl/nieuws/6001041/datalek-gemeente-eindhoven-deelt-persoonsgegevens-met-openbare-ai-websites","primarySourceLabel":"Omroep Brabant, 19 December 2025: Eindhoven data breach, municipality shares personal data with public AI websites","firstPublishedAt":"2026-10-12T03:35:14.550754+00:00","updatedAt":"2026-10-12T03:35:14.550754+00:00","scopeVersion":"facts-v3","tags":["privacy","government","municipality","youth-care","health-data","chatgpt","unauthorised-ai-use","netherlands"]},{"id":"2025-alsace-franchise-recruiter-chatgpt-cv-filter-age-sex-origin-criteria-ex-employees-allege","caseFacts":{"claims":[{"id":"c1","status":"reported","evidence":[{"locator":"'quelque 740.000 CV de commerciaux. Chiffre ensuite ramené à 130.000 en seulement cinq jours, grâce à ChatGTP, après application de filtres discriminants, raconte un autre ex-salarié.'","relation":"supports","source_id":"s1"}],"assertion":"A former employee says the company collected about 740,000 sales CVs from job platforms in spring 2025 and reduced them to 130,000 in five days with ChatGPT after applying discriminatory filters.","causal_attribution":"A former employee's account relayed by AFP."},{"id":"c2","status":"reported","evidence":[{"locator":"'et à exclure les moins de 27 ans.' and 'il faut (que) le nom de famille du candidat ne SOIT PAS d' and 'un motif de rejet figure en face de chaque nom: \"Hors cible - trop jeune\", ou encore \"Trop âgé\". Les femmes, repérées à leur prénom, sont renvoyées à leurs chères études avec la mention \"Exception\".'","relation":"supports","source_id":"s1"},{"locator":"'un motif de rejet figure en face de chaque nom'","relation":"supports","source_id":"s2"}],"assertion":"AFP consulted the prompt, which told the AI to exclude candidates over 55 or under 27 and those whose surname was of African origin, and a results table assigning a rejection reason to each name, with women marked \"Exception\".","causal_attribution":"Documents described by AFP; the record does not reproduce them and AFP does not say how many candidates the table covers."},{"id":"c3","status":"reported","evidence":[{"locator":"'tous les commerciaux retenus ont le même profil: blanc, dans la quarantaine'","relation":"supports","source_id":"s1"}],"assertion":"A former employee said all the sales agents retained had the same profile, white and in their forties.","causal_attribution":"A former employee's statement."},{"id":"c4","status":"reported","evidence":[{"locator":"'contenait des développements contraires à notre cahier des charges et aux consignes de la direction' and 'a jamais été mise en service et n' and 'a servi à évaluer aucun candidat'","relation":"supports","source_id":"s1"}],"assertion":"The company denies asking anyone to build discriminatory criteria into its tools; it says a version containing developments contrary to its specifications was never put into service, evaluated no candidate and was built by a since-dismissed former employee.","causal_attribution":"The company's statement to AFP, recorded as contrary evidence."},{"id":"c5","status":"reported","evidence":[{"locator":"'Le troisième affirme avoir saisi la Commission nationale de l' and 'ainsi que le défenseur des droits' and 'a indiqué ne pas pouvoir communiquer sur un cas individuel.'","relation":"supports","source_id":"s1"}],"assertion":"One former employee says he has referred the matter to the CNIL and the Défenseur des droits; the CNIL said it could not comment on an individual case.","causal_attribution":"Former employee's statement and the CNIL response, via AFP."},{"id":"c6","status":"reported","evidence":[{"locator":"'Plus que pro a trié des milliers de CV sur des critères de genre et d' and 'entre 2023 et 2026, Plus que pro a mis en place un système de discrimination pour choisir ses futurs franchisés, basé sur l'","relation":"supports","source_id":"s3"}],"assertion":"Rue89 Strasbourg reported, from working documents and testimony, that the company sorted thousands of CVs on criteria of gender and supposed origin without the candidates' knowledge, between 2023 and 2026.","causal_attribution":"A separate investigation; its readable portion does not mention AI, so it supports the discriminatory sorting and not the AI role."}],"effects":[{"label":"job-seekers screened out of consideration by an AI filter applying age, sex and surname-origin criteria, according to former employees","claim_id":"c2","direction":"negative"}],"sources":[{"id":"s1","url":"https://www.boursorama.com/actualite-economique/actualites-amp/ni-femmes-ni-profils-exotiques-quand-l-ia-est-programmee-pour-discriminer-8819657638628fd90de0e90be62ff869","kind":"news_report","access":"read","language":"fr","translation_note":"Read in French by the research agent (an AI) on 2026-10-11; quotations in the record are the original French and the English renderings in public text are the research agent's own translation. No human translator or reviewer inspected the text. AFP dispatch on Boursorama (AMP page), HTTP 200, full text retained.","independence_group":"afp-plus-que-pro"},{"id":"s2","url":"https://www.20min.ch/fr/story/alsace-ni-femmes-ni-noms-africains-l-ia-fait-le-job-pour-une-start-up-103647007","kind":"news_report","access":"read","language":"fr","translation_note":"Read in French by the research agent (an AI) on 2026-10-11; quotations in the record are the original French and the English renderings in public text are the research agent's own translation. No human translator or reviewer inspected the text. 20 minutes republishes the AFP dispatch and credits AFP; HTTP 200.","independence_group":"afp-plus-que-pro"},{"id":"s3","url":"https://www.rue89strasbourg.com/discrimination-recrutement-plus-que-pro-398436","kind":"news_report","access":"read","language":"fr","translation_note":"Read in French by the research agent (an AI) on 2026-10-11; quotations in the record are the original French and the English renderings in public text are the research agent's own translation. No human translator or reviewer inspected the text. Subscriber-only investigation: only the standfirst and opening paragraphs were readable (HTTP 200); the rest, including any AI detail, was not inspected.","independence_group":"rue89-strasbourg-plus-que-pro"}],"version":1,"ai_roles":["institutional_use"],"contexts":["work"],"unknowns":["How many candidates were excluded on discriminatory criteria by the AI filter, and whether the filter ran on the collected CVs as the former employees say or never ran as the company says.","Whether the Rue89 Strasbourg investigation ties the sorting system (named Vortex) to the AI tool; its subscriber-only text was not read in full.","The outcome of the CNIL and Défenseur des droits referrals."],"geography":{"basis":"AFP places the company and the former employees who ran the screening in Schiltigheim, near Strasbourg, France. The candidates' countries are not stated; the CVs came from job platforms such as Hellowork. No court proceeding is reported.","court_countries":[],"event_countries":["FR"],"affected_person_countries":[]},"publication":{"basis":"Published as an automated hiring decision case: AFP describes a prompt instructing an AI to reject candidates by age and surname origin and a results table of per-candidate rejection reasons, and a former employee names ChatGPT as the filter that cut 740,000 CVs to 130,000. Rue89 Strasbourg separately reports discriminatory CV sorting at the company. The company denies that the discriminatory version was used on any candidate, so causation is recorded as disputed and all claims as reported.","reviewed_on":"2026-10-11"},"ai_involvement":{"basis":"AFP reports that it consulted the prompt sent to the AI, which instructed it to estimate candidates' ages and exclude those over 55 or under 27 and to require that surnames not be of African origin, and a results table assigning a rejection reason to each name. A former employee says the CVs were cut from 740,000 to 130,000 in five days \"grâce à ChatGTP\" after the discriminatory filters were applied. The AI output (the per-candidate rejection decisions) is the described connection to the candidates' exclusion. The company says the discriminatory version was never used on any candidate.","status":"reported"},"person_relations":["made_decision_about"]},"name":"France: ex-employees say Alsace firm Plus que pro used AI to screen out job-seekers' CVs by age, sex and African-origin surnames; firm says it never ran","summary":"Plus que pro, a company founded in 2014 in Schiltigheim near Strasbourg that recruits franchised sales agents, is accused by three former employees interviewed by AFP of building a recruitment tool that screened candidates on unlawful criteria with the help of AI. According to one of them, in spring 2025 the company collected about 740,000 sales CVs from job platforms such as Hellowork and reduced them to 130,000 in five days with ChatGPT after applying discriminatory filters. AFP says it consulted the prompt, which told the AI to estimate each candidate's age, keep those under 55, infer age from the baccalaureate year and exclude those under 27, and require that the candidate's surname not be of African origin, and a results table listing a rejection reason against each name, such as \"too young\" or \"too old\", with women identified by first name marked \"Exception\". A former employee said that all the sales agents retained had the same profile, white and in their forties; another said the AI only systematised a practice already established when CVs were sorted by hand. One former employee says he has referred the matter to the CNIL and the Défenseur des droits; the CNIL said it could not comment on an individual case. The company denies ever asking anyone to build discriminatory criteria into its tools; it acknowledges that one version of the tool contained developments contrary to its specifications, but says that version was never put into service, evaluated no candidate and was developed by a former employee acting on their own initiative, who has since been dismissed. Rue89 Strasbourg separately reported, from working documents and testimony, that the company sorted thousands of CVs on criteria of gender and supposed origin without the candidates' knowledge between 2023 and 2026; its subscriber-only report could be read only in part.","incidentDate":"2025-01-01","incidentKind":"bounded_series","incidentDatePrecision":"year","exposurePattern":"unknown","reportedDate":"2026-10-09","aiSystem":"A recruitment tool built from late 2024 by the company, which sent instructions to a generative AI to screen collected CVs; a former employee quoted by AFP names ChatGPT (\"ChatGTP\" in the AFP text).","aiProduct":"ChatGPT (reported)","severity":"low","verificationStatus":"credible","harmCategories":[],"harmOutcomes":["professional_harm"],"harmOutcomeSummary":"According to former employees and documents described by AFP, job-seekers whose CVs were collected from job platforms were screened out of consideration for franchised sales-agent positions by an AI filter applying age, sex and surname-origin criteria. No individual rejected candidate is identified in the inspected AI-related reporting and the number excluded on those criteria is not reported. The company says the discriminatory version was never used on any candidate.","frameworkFacets":[],"causationStatus":"disputed","participantUsersAffectedMin":0,"otherPeopleHarmedMin":0,"affectedCountStatus":"unquantified","affectedCountEvidence":"No reliable count of people harmed. The 740,000 and 130,000 figures are CVs collected and retained, as described by a former employee; the difference includes candidates removed for any reason and is not a count of people excluded on discriminatory criteria. Numeric fields are zero placeholders.","victimAgeRange":"adult","jurisdiction":"FR","platformType":"other","outcomeType":"media_coverage","outcomeStatus":"ongoing","primarySourceUrl":"https://www.boursorama.com/actualite-economique/actualites-amp/ni-femmes-ni-profils-exotiques-quand-l-ia-est-programmee-pour-discriminer-8819657638628fd90de0e90be62ff869","primarySourceLabel":"AFP via Boursorama, 9 October 2026: an Alsace start-up accused by ex-employees of building a discriminatory AI recruitment tool","firstPublishedAt":"2026-10-11T03:25:30.958981+00:00","updatedAt":"2026-10-11T03:25:30.958981+00:00","scopeVersion":"facts-v3","tags":["hiring","recruitment","discrimination","chatgpt","cv-screening","france","alsace","made-decision-about","institutional-use"]},{"id":"2025-uk-upper-tribunal-asylum-appeal-chatgpt-grounds-fake-cases-invented-quotes-sra-self-report","caseFacts":{"claims":[{"id":"c1","status":"documented","evidence":[{"locator":"the grounds of appeal were completely infected by the use of AI. In addition to the citation of fake cases, and paragraphs from cases which are not contained in those judgments, the Grounds of Appeal also quote paragraphs from the decision of the First-tier Tribunal which do not appear in that decision.","relation":"supports","source_id":"s1"},{"locator":"This was pure invention, which was not to be found anywhere in Judge Balroop’s decision, yet criticism of it formed the basis for one aspect of ground (3) in the grounds of appeal.","relation":"supports","source_id":"s1"},{"locator":"In this case, Mr Badar identified at least 18 separate errors in the grounds of appeal and withdrew three of the four grounds.","relation":"supports","source_id":"s1"}],"assertion":"The appellant's then solicitors drafted grounds of appeal that her counsel accepted were completely infected by the use of AI, citing fake cases and paragraphs not contained in the cited judgments and quoting passages from the First-tier Tribunal decision that do not appear in it; the tribunal found one purported quotation to be pure invention.","causal_attribution":"Established by the tribunal's own findings and its record of counsel's concession in the published decision. The Gazette repeats the decision and adds no independent origin."},{"id":"c2","status":"documented","evidence":[{"locator":"Both were produced by a junior member of staff, namely a paralegal, using ChatGPT, which itself raises significant concerns about client confidentiality and legal privilege.","relation":"supports","source_id":"s1"},{"locator":"This occurred because Sarker Solicitors did not check the accuracy of the grounds of appeal, did not consider the grant of permission and did not check the accuracy of the skeleton argument.","relation":"supports","source_id":"s1"},{"locator":"They accepted that AI was used to generate the grounds of appeal filed in the FtT and the skeleton argument filed in the UT.","relation":"supports","source_id":"s1"},{"locator":"The grounds of appeal and skeleton argument were drafted by an unnamed paralegal working for east London firm Sarker Solicitors.","relation":"supports","source_id":"s2"}],"assertion":"The tribunal found that both the grounds of appeal and the skeleton argument were produced by a junior member of staff, a paralegal, using ChatGPT, and that the firm did not check their accuracy; the solicitors had accepted that AI was used to generate both documents.","causal_attribution":"The product identification and the failure to check are the tribunal's findings in the decision, recorded with the solicitors' acceptance that AI generated the documents. The Gazette's statement that the drafter was an unnamed paralegal relays the decision (same independence group)."},{"id":"c3","status":"documented","evidence":[{"locator":"A hearing was listed before Deputy Upper Tribunal Judge Hobbs on 17 September 2025 to consider whether Judge Balroop had erred in law.","relation":"supports","source_id":"s1"},{"locator":"Judge Hobbs adjourned the hearing, with directions which included the following:","relation":"supports","source_id":"s1"},{"locator":"We note that issues (2) to (4), which occupied a great deal of time at the hearing before us, only arose because of Sarker Solicitors’ inappropriate use of AI.","relation":"supports","source_id":"s1"},{"locator":"The appellant withdrew the original grounds (1), (2) and (4).","relation":"supports","source_id":"s1"},{"locator":"The appellant’s application for permission to amend ground (3) is granted.","relation":"supports","source_id":"s1"},{"locator":"We accept Mr Terrell’s submission that it is not in the interests of justice in this case that the appellant be permitted to advance entirely new grounds, or to replace grounds riddled with misrepresentations, well beyond the original time limit for seeking permission to appeal.","relation":"supports","source_id":"s1"},{"locator":"We therefore refuse the application for permission to amend in respect of the new grounds (4) and (5).","relation":"supports","source_id":"s1"},{"locator":"The appellant’s application for permission to amend grounds (4) and (5) is refused. For completeness, grounds (4) and (5) have been considered in the alternative and disclose no error of law.","relation":"supports","source_id":"s1"},{"locator":"However, ground (3) discloses no error of law.","relation":"supports","source_id":"s1"},{"locator":"The appellant’s appeal therefore fails.","relation":"supports","source_id":"s1"},{"locator":"The judge declined to set aside permission to appeal but dismissed the appeal in any case. The Home Office was invited to file and serve any wasted costs application on Sarker Solicitors within 28 days.","relation":"supports","source_id":"s2"}],"assertion":"The error-of-law hearing listed for 17 September 2025 was adjourned, and three of the five issues at the final hearing, which occupied a great deal of its time, arose only because of the solicitors' use of AI. The appellant withdrew three of her four original grounds. The tribunal allowed an amended version of her surviving ground (3) but found that it disclosed no error of law. It refused permission to add new grounds (4) and (5), holding that it was not in the interests of justice to let her advance entirely new grounds, or replace grounds riddled with misrepresentations, well beyond the original time limit, and in the alternative found those grounds without merit. Her appeal failed.","causal_attribution":"Procedural history and disposition are directly established by the decision. The decision attributes the adjournment and the wasted hearing time to the solicitors' misuse of AI. It gives the misrepresentation-riddled original grounds and the late application as its reasons for refusing the new grounds (para 65), and it also found those grounds without merit in the alternative (para 74). It states separately that the surviving ground failed on its merits, so the dismissal itself is not attributed to the AI output."},{"id":"c4","status":"documented","evidence":[{"locator":"Two individuals have reported themselves to the Solicitors Regulation Authority and so we will say nothing about the professional conduct issues involved.","relation":"supports","source_id":"s1"},{"locator":"The respondent shall file and serve on Sarker Solicitors any wasted costs application, or confirmation that costs are not being pursued, within 28 days of this decision being sent to the parties.","relation":"supports","source_id":"s1"},{"locator":"Two individuals involved in filing court documents using ChatGPT which were riddled with mistakes and invented quotes have reported themselves to the Solicitors Regulation Authority.","relation":"supports","source_id":"s2"}],"assertion":"The decision states that two individuals have reported themselves to the Solicitors Regulation Authority, and directs the Home Office to file any wasted costs application against the firm within 28 days; no costs decision has been taken.","causal_attribution":"The self-reports and the costs directions are stated in the decision itself. The Gazette repeats them from the decision (same independence group), so the claim is documented, not corroborated. Whether the two individuals are the paralegal and the supervising solicitor is not stated."},{"id":"c5","status":"documented","evidence":[{"locator":"It would be misleading to describe them as having been caused by AI, rather than by the solicitors’ misuse of AI.","relation":"supports","source_id":"s1"},{"locator":"This was one of the more extreme examples of the misuse of AI, rather than a mere mis-citation of particular cases.","relation":"supports","source_id":"s1"}],"assertion":"The tribunal said it would be misleading to describe the irregularities as caused by AI rather than by the solicitors' misuse of AI. It recorded the Home Office's submission that this was one of the more extreme examples of the misuse of AI.","causal_attribution":"The tribunal's own qualification, preserved as contrary evidence on causation: the decision places responsibility with the solicitors. The Home Office's characterisations (at best negligent, at worst reckless, and one of the more extreme examples of the misuse of AI) are relayed in the decision as its submissions (para 43)."}],"effects":[{"label":"Appellant's protection appeal pursued on AI-generated grounds containing fake cases and invented quotations; hearing adjourned and three of four grounds withdrawn","claim_id":"c3","direction":"negative"},{"label":"Two individuals self-reported to the Solicitors Regulation Authority; wasted costs application against the firm invited","claim_id":"c4","direction":"negative"}],"sources":[{"id":"s1","url":"https://tribunalsdecisions.service.gov.uk/utiac/ui-2025-002601","kind":"court_record","access":"read","language":"en","translation_note":"","independence_group":"ut-decision-ui-2025-002601"},{"id":"s2","url":"https://www.lawgazette.co.uk/news/duo-self-report-to-sra-over-grounds-of-appeal-infected-with-ai/5128148.article","kind":"news_report","access":"read","language":"en","translation_note":"","independence_group":"ut-decision-ui-2025-002601"}],"version":1,"ai_roles":["others_use"],"contexts":["justice"],"unknowns":["The date on which the paralegal generated the grounds of appeal is not stated; it falls between the First-tier decision of 23 April 2025 and the hearing listed for 17 September 2025.","Whether the two individuals who self-reported to the Solicitors Regulation Authority are the paralegal and the supervising solicitor is not stated, and the outcome of any SRA consideration is unknown.","Whether the Home Office pursued wasted costs, and any amount, was not decided in the published decision.","What prompts were given to ChatGPT, and whether client material was entered into it, is not described; the tribunal only notes that the use raises confidentiality and privilege concerns.","Whether the appellant's appeal would have succeeded on properly drafted grounds cannot be known. The tribunal found that her surviving, amended ground disclosed no error of law, and that the two new grounds it refused to admit disclosed none either when considered in the alternative."],"geography":{"basis":"The decision is an Upper Tribunal (Immigration and Asylum Chamber) decision heard at Field House, London, in an appeal against a UK Home Office refusal; the solicitors are described by the Gazette as an east London firm. The appellant is a Namibian national whose appeal concerns her removal from the UK, so her country is recorded as GB on the basis of the proceedings in which she was harmed, and Namibia is not recorded because no harm occurred there.","court_countries":["GB"],"event_countries":["GB"],"affected_person_countries":["GB"]},"publication":{"basis":"The published Upper Tribunal decision (tribunalsdecisions.service.gov.uk, UI-2025-002601) and the Law Society Gazette report of 8 October 2026 were read in full on 2026-10-09, and the decision was re-read on 2026-10-11 for this correction. The decision records what ChatGPT generated (fake cases, misattributed paragraphs, invented quotations from the First-tier judge) and ties that output to the adjourned hearing, the wasted hearing time, the withdrawn grounds, the refusal of permission to add new grounds, the regulator self-reports and the costs directions. The appellant is anonymised by tribunal order and is described only by nationality and appeal type; the firm is named because the operative costs directions are addressed to it; the paralegal is unnamed. The tribunal's statement that the solicitors' misuse, not AI, caused the irregularities is preserved.","reviewed_on":"2026-10-11"},"ai_involvement":{"basis":"The Upper Tribunal's published decision finds that the grounds of appeal and the skeleton argument filed on the appellant's behalf were produced by a paralegal using ChatGPT, and describes what that output contained: citations of fake cases, paragraphs not found in the cited judgments, and purported quotations from the First-tier judge's decision that the tribunal calls pure invention. The decision connects that output to the harm: the September 2025 hearing was adjourned after the Home Office queried the citation errors, three of the five issues at the final hearing, which occupied a great deal of its time, arose only because of the solicitors' use of AI, the appellant withdrew three of four grounds, and the tribunal invited a wasted costs application and noted two self-reports to the regulator. The FtT judge granted permission on those grounds while noting the misquotations, and the Upper Tribunal's decisions not to set aside permission and to refuse the new grounds were taken in the express context of the misrepresentations (paras 46-47, 65 and 74). The tribunal itself states that the irregularities were caused by the solicitors' misuse of AI rather than by AI, and the decision records the solicitors' own acceptance that AI was used. The firm's acceptance and the tribunal's findings come from one record, so the status is supported rather than corroborated by an independent source.","status":"supported"},"person_relations":["acted_on_behalf","informed_decision_about"]},"name":"UK Upper Tribunal: solicitors' ChatGPT-drafted grounds in an asylum appeal cited fake cases and invented quotes; two individuals self-report to the SRA","summary":"A published Upper Tribunal (Immigration and Asylum Chamber) decision issued on 21 September 2026 records that the grounds of appeal and skeleton argument in a Namibian national's protection appeal were produced by a paralegal at her then solicitors using ChatGPT. The tribunal recorded her counsel's acceptance that the grounds cited fake cases, quoted passages absent from those judgments and attributed to the First-tier judge sentences he never wrote, and itself called one purported quotation pure invention. The September 2025 error-of-law hearing was adjourned, and the appellant later withdrew three of her four original grounds. The tribunal allowed an amended version of her one surviving ground but found that it disclosed no error of law. It refused her permission to add two new grounds, holding that she should not be allowed to advance entirely new grounds, or replace grounds riddled with misrepresentations, so long after the time limit, and in the alternative found those grounds without merit. It directed the Home Office to file any wasted costs application against the firm. The decision states that two individuals have reported themselves to the Solicitors Regulation Authority; the Law Society Gazette reported the decision on 8 October 2026.","incidentDate":"2025-09-17","incidentEndDate":"2026-09-21","incidentKind":"bounded_series","incidentDatePrecision":"range","exposurePattern":"unknown","reportedDate":"2026-10-06","aiSystem":"ChatGPT, used by a paralegal to generate grounds of appeal and a skeleton argument (as found in the Upper Tribunal decision)","aiProduct":"ChatGPT","aiCompany":"OpenAI","severity":"low","verificationStatus":"credible","harmCategories":[],"harmOutcomes":["legal_harm","professional_harm"],"harmOutcomeSummary":"The tribunal records that the appellant's error-of-law hearing in September 2025 was adjourned, that three of the five issues at the eventual hearing, which occupied a great deal of its time, arose only because of her solicitors' inappropriate use of AI, that she withdrew three of her four original grounds, that it refused her permission to add two new grounds, holding that she should not be allowed to advance entirely new grounds, or replace grounds riddled with misrepresentations, long after the time limit, and that her appeal fails. It found her one surviving, amended ground disclosed no error of law and, in the alternative, that the two refused grounds disclosed none either. The decision states that two individuals have reported themselves to the Solicitors Regulation Authority and directs the Home Office to file any wasted costs application against the firm; no costs decision has been taken.","frameworkFacets":[],"causationStatus":"established","participantUsersAffectedMin":0,"otherPeopleHarmedMin":1,"affectedCountStatus":"documented_minimum","affectedCountEvidence":"One person counted: the appellant, whose appeal proceeded on AI-generated grounds she did not draft (the decision describes her as the appellant throughout). The paralegal who used ChatGPT and the supervising solicitor face professional consequences of their own conduct (self-report to the SRA, possible wasted costs) and are not counted as harmed persons; the two self-reporting individuals are not named as the paralegal and the solicitor in the decision.","victimAgeRange":"adult","jurisdiction":"GB","platformType":"chatbot","primarySourceUrl":"https://tribunalsdecisions.service.gov.uk/utiac/ui-2025-002601","primarySourceLabel":"Upper Tribunal (Immigration and Asylum Chamber), UI-2025-002601, decision issued 21 September 2026, published 6 October 2026","firstPublishedAt":"2026-10-09T03:28:56.311373+00:00","updatedAt":"2026-10-11T07:43:28.275795+00:00","scopeVersion":"facts-v3","tags":[]},{"id":"2025-us-neh-humanities-grant-mass-termination-chatgpt-dei-screening","caseFacts":{"claims":[{"id":"c1","status":"documented","evidence":[{"locator":"The Mass Termination is DECLARED unlawful,","relation":"supports","source_id":"s1"},{"locator":"was unlawful because it was undertaken in violation of the First Amendment","relation":"supports","source_id":"s1"},{"locator":"the resulting terminations were ultra vires","relation":"supports","source_id":"s1"},{"locator":"the termination, in early April 2025, of more than 1,400 grants,","relation":"supports","source_id":"s1"},{"locator":"is PERMANENTLY ENJOINED from enforcing, implementing, or giving any effect to the Mass Termination.","relation":"supports","source_id":"s1"}],"assertion":"On 7 May 2026 the court declared the April 2025 mass termination of NEH grants (more than 1,400 grants, over $100 million) unlawful, unconstitutional and ultra vires, and permanently enjoined its enforcement.","causal_attribution":"The declaration rests on the First Amendment, equal protection and lack of statutory authority. It does not rest on a finding that ChatGPT was inaccurate."},{"id":"c2","status":"documented","evidence":[{"locator":"Does the following relate at all to DEI? Respond factually in less than 120 characters. Begin with ‘Yes.’ or ‘No.’ followed by a brief explanation.","relation":"supports","source_id":"s1"},{"locator":"did not provide ChatGPT with a definition of “DEI”","relation":"supports","source_id":"s1"},{"locator":"it enlisted the help of ChatGPT.","relation":"supports","source_id":"s2"},{"locator":"two employees of the Department of Government Efficiency (DOGE) used ChatGPT to determine whether previously approved National Endowment for the Humanities (NEH) grants should be canceled","relation":"context","source_id":"s4"}],"assertion":"A DOGE staff member used ChatGPT with a standardized prompt asking whether each grant description related at all to DEI, without defining DEI for the model.","causal_attribution":"The opinion recites deposition testimony and the spreadsheet. The prompt text is a quotation of the court record."},{"id":"c3","status":"documented","evidence":[{"locator":"thereby combining DOGE’s AI-generated classifications with NEH staff recommendations.","relation":"supports","source_id":"s1"},{"locator":"did not know ChatGPT had been used","relation":"supports","source_id":"s1"}],"assertion":"The ChatGPT outputs were combined with NEH staff ratings into the lists used to select grants for termination, and the NEH acting chair testified to not knowing ChatGPT had been used.","causal_attribution":"The court ties the ChatGPT-generated rationales to the lists presented to the NEH chair. The acting chair's account is deposition testimony as recited by the court."},{"id":"c4","status":"documented","evidence":[{"locator":"were terminated on that basis and that basis alone.","relation":"supports","source_id":"s1"},{"locator":"Of the 1,163 grant proposals DOGE analyzed via ChatGPT for DEI-related content, 1,057 were flagged, and just 42 were kept","relation":"context","source_id":"s3"}],"assertion":"The opinion states that 1,057 grants were identified as DEI-related through the DOGE review process and terminated on that basis alone.","causal_attribution":"The opinion attributes the identification to a process that combined keyword searches and ChatGPT rationales. The opinion says ChatGPT rationales were generated in a second-stage review of grants NEH staff had labelled \"N/A\" and were merged with NEH ratings. It does not apportion the 1,057 between NEH ratings and ChatGPT. Fortune attributes all 1,057 flags to ChatGPT from a spreadsheet exhibit."},{"id":"c5","status":"documented","evidence":[{"locator":"Yes. This description highlights how the plastics industry influenced municipal government officials to favor plastics","relation":"supports","source_id":"s1"},{"locator":"Nevertheless, ChatGPT classified the project as “DEI,”","relation":"supports","source_id":"s1"},{"locator":"The utter lack of reasoning behind so many of its “rationales”","relation":"supports","source_id":"s1"}],"assertion":"The court describes ChatGPT classifying projects as DEI on the basis of short descriptions, including a project on the plastics industry and municipal building codes, with rationales the court found lacking.","causal_attribution":"The court characterises the outputs. Whether each terminated grant was flagged by ChatGPT is not stated for the individual plaintiffs."},{"id":"c6","status":"reported","evidence":[{"locator":"The individual plaintiffs are scholars and writers who received NEH grants for specific research and writing projects and whose awards were terminated before the completion of their funded work.","relation":"supports","source_id":"s1"},{"locator":"allege concrete professional, financial, and scholarly harms","relation":"supports","source_id":"s1"},{"locator":"That funding was terminated effective April 2, 2025, resulting in the loss of the fellowship itself.","relation":"supports","source_id":"s1"},{"locator":"individual grantees whose awards were terminated suffered a concrete economic injury sufficient to confer Article III standing.","relation":"supports","source_id":"s1"}],"assertion":"Seven individual grantees whose NEH awards were terminated before completion allege professional, financial and scholarly harm, including lost funding, and in one case loss of a fellowship that depended on NEH funding.","causal_attribution":"The harms are the plaintiffs' allegations as summarised by the court, and the standing holding is recited from the court's earlier preliminary injunction opinion. The link between ChatGPT and the termination of these seven specific awards is not stated in the inspected passages."},{"id":"c7","status":"documented","evidence":[{"locator":"DOGE officials lacked statutory authority to identify, select, or direct the termination of NEH grants","relation":"supports","source_id":"s1"}],"assertion":"The court declared that DOGE officials lacked statutory authority to identify, select or direct the termination of NEH grants.","causal_attribution":"A legal declaration in the judgment."},{"id":"c8","status":"reported","evidence":[{"locator":"Yes. Improving HVAC systems enhances preservation conditions for collections, aligning with the goal of providing greater access to diverse audiences. #DEI,","relation":"supports","source_id":"s3"},{"locator":"We were able to recoup about 70% of the original award through the grant termination clause,","relation":"supports","source_id":"s3"}],"assertion":"A North Carolina history museum's $349,000 HVAC grant was flagged by ChatGPT as DEI and terminated after the project had begun, and the museum director says it recouped about 70% of the award through the termination clause.","causal_attribution":"Organisation-level example from one outlet and one interview. It is not counted as harm to a person."}],"effects":[{"label":"Individual grantees lost or had interrupted NEH funding when their awards were terminated in a mass termination that relied on a ChatGPT DEI screen","claim_id":"c6","direction":"negative"}],"sources":[{"id":"s1","url":"https://www.nysd.uscourts.gov/sites/default/files/2026-05/NEH.sj_.pdf","kind":"court_record","access":"read","language":"en","translation_note":"","independence_group":"sdny-opinion-2026-05"},{"id":"s2","url":"https://www.insidehighered.com/news/faculty-issues/research/2026/03/11/how-doge-gutted-neh-22-days","kind":"news","access":"read","language":"en","translation_note":"","independence_group":"neh-litigation-record-news"},{"id":"s3","url":"https://fortune.com/2026/03/19/doge-cancelled-350000-hvac-grant-dei-lawsuit-elon-musk/","kind":"news","access":"read","language":"en","translation_note":"","independence_group":"neh-litigation-record-news"},{"id":"s4","url":"https://www.artforum.com/news/doge-allegedly-used-chatgpt-to-cancel-humanities-grants-1234745040/","kind":"news","access":"read","language":"en","translation_note":"","independence_group":"neh-litigation-record-news"}],"version":1,"ai_roles":["institutional_use"],"contexts":["work","education","public_services"],"unknowns":["Whether ChatGPT reviewed or flagged the specific grants of the seven individual plaintiffs is not stated in the passages inspected.","The dates on which ChatGPT was used are not stated beyond the second stage of review beginning on 12 March 2025 and terminations sent 1 to 3 April 2025. The incident dates are the bounded series between those points.","The number of individual people (as opposed to organizations) whose awards were terminated is not stated.","The opinion separates grants marked \"DEI\" from grants terminated without a DEI mark (its Biden-Era Subclass). It does not say which group any of the seven individual plaintiffs' grants belonged to.","Whether funding has been restored is unknown. The judgment states it does not require immediate payment.","Appeal status of the May 2026 judgment is unknown.","The New York Times article listed in the AIID entry returned HTTP 403 and was not read. The opinion is 143 pages and was read through its factual sections, the standing recital and the judgment, not every legal section."],"geography":{"basis":"The opinion describes terminations by the US federal agency NEH and by DOGE staff of grants to scholars, writers and institutions, and describes the individual plaintiffs as holding positions at US universities. The court is a US federal district court (S.D.N.Y.).","court_countries":["US"],"event_countries":["US"],"affected_person_countries":["US"]},"publication":{"basis":"A federal court opinion (7 May 2026), Inside Higher Ed and Fortune were read. Claims about the court record are documented by the opinion. Individual grantee harms are the plaintiffs' allegations as summarised by the court. Individual plaintiffs and DOGE staff are not named in public fields.","reviewed_on":"2026-09-29"},"ai_involvement":{"basis":"The federal court opinion finds that a DOGE staff member submitted grant descriptions to ChatGPT with a DEI yes/no prompt and that the outputs were folded into the lists used to select grants for termination. The opinion does not state whether each of the seven individual plaintiffs' grants was reviewed or flagged by ChatGPT.","status":"supported"},"person_relations":["made_claim_about"]},"name":"US: National Endowment for the Humanities grants terminated after DOGE staff used ChatGPT to flag projects as DEI-related","summary":"In March and early April 2025 (the opinion gives no date for the ChatGPT step), a DOGE staff member working with the National Endowment for the Humanities (NEH) submitted short grant descriptions to ChatGPT with a prompt asking whether each project related at all to DEI, and the answers were combined with NEH staff ratings into lists of grants to terminate. More than 1,400 NEH grants (over $100 million) were terminated in notices sent 1 to 3 April 2025. A federal district court found the mass termination unlawful in May 2026 and described the ChatGPT step in its findings. Seven individual grantees who sued report lost or interrupted funding for their research and writing projects. The opinion does not state whether ChatGPT reviewed those seven grants.","incidentDate":"2025-03-12","incidentEndDate":"2025-04-03","incidentKind":"bounded_series","incidentDatePrecision":"day","exposurePattern":"unknown","reportedDate":"2026-03-08","aiSystem":"ChatGPT (OpenAI), used through a standardized DEI yes/no prompt","aiProduct":"ChatGPT","aiCompany":"OpenAI","severity":"medium","verificationStatus":"credible","harmCategories":[],"harmOutcomes":["financial_loss","professional_harm"],"harmOutcomeSummary":"The court opinion describes seven individual grantee plaintiffs whose NEH awards were terminated before completion, and reports their allegations of lost funding, interrupted research and leave from employment taken in reliance on the awards. Individual counts beyond these seven are not established in the inspected sources.","frameworkFacets":[],"causationStatus":"unclear","participantUsersAffectedMin":0,"otherPeopleHarmedMin":7,"affectedCountStatus":"partial","affectedCountEvidence":"The court opinion names seven individual grantee plaintiffs whose awards were terminated. It states the terminations reached more than 1,400 grants awarded to scholars, writers, research institutions and other humanities organizations, and does not give a count of individual people. The opinion does not state whether these seven grants were flagged by ChatGPT. Organizations, members of plaintiff associations and the wider class are not counted.","victimAgeRange":"adult","platformType":"assistant","primarySourceUrl":"https://www.nysd.uscourts.gov/sites/default/files/2026-05/NEH.sj_.pdf","primarySourceLabel":"S.D.N.Y. opinion and order granting summary judgment, 7 May 2026","firstPublishedAt":"2026-09-29T13:28:29.039734+00:00","updatedAt":"2026-09-30T01:17:19.404135+00:00","scopeVersion":"facts-v3","tags":["historical-2025"]},{"id":"2025-krafton-unknown-worlds-executives-terminated-after-chatgpt-takeover-strategy","caseFacts":{"claims":[{"id":"c1","status":"documented","evidence":[{"locator":"Krafton breached the EPA by terminating the Key Employees without valid Cause and by improperly seizing operational control of Unknown Worlds.","relation":"supports","source_id":"s1"},{"locator":"Date Decided: March 16, 2026","relation":"supports","source_id":"s1"},{"locator":"Krafton agreed the studio would remain independent and that its leadership would retain operational control and could only be fired for cause, according to the ruling","relation":"supports","source_id":"s2"}],"assertion":"On 16 March 2026 the Delaware Court of Chancery entered judgment that Krafton breached the acquisition agreement by terminating the three Key Employees without valid Cause and by improperly seizing operational control of the studio.","causal_attribution":"Phase One judgment on the contract claims. The court reserves damages for a second phase and does not decide whether the earnout was wrongfully impaired."},{"id":"c2","status":"documented","evidence":[{"locator":"a “dismissal with cause” would not eliminate the earnout obligation","relation":"supports","source_id":"s1"},{"locator":"turned to ChatGPT for help.","relation":"supports","source_id":"s1"},{"locator":"Now, chatgpt starts to answer that it is difficult to cancel the earn-out.","relation":"supports","source_id":"s1"},{"locator":"When the AI chatbot responded that the earnout would be “difficult to cancel,”","relation":"supports","source_id":"s1"}],"assertion":"The court found that after the Head of Corporate Development at Krafton warned the CEO that a dismissal with cause would not eliminate the earnout obligation, the CEO turned to ChatGPT, and that the CEO told the Head of Corporate Development in a Slack message that ChatGPT answered the earnout would be difficult to cancel.","causal_attribution":"Findings of fact drawn from Slack messages and trial testimony. The chat itself was deleted."},{"id":"c3","status":"documented","evidence":[{"locator":"The AI chatbot prepared a “Response Strategy to a ‘No-Deal’ Scenario,”","relation":"supports","source_id":"s1"},{"locator":"Lock down Steam/console publishing rights and access rights over code/build pipeline","relation":"supports","source_id":"s1"},{"locator":"Preemptive Framing - Repeat that protecting quality and fan trust is the highest priority","relation":"supports","source_id":"s1"},{"locator":"Systematic materials for legal defense - Prepare contract interpretation memorandums, log all communications","relation":"supports","source_id":"s1"},{"locator":"Team retention - Operation of retention packages for key personnel and rapid backfill pipelines","relation":"supports","source_id":"s1"},{"locator":"Over the next month, Krafton followed most of ChatGPT’s recommendations.","relation":"supports","source_id":"s1"},{"locator":"Over the next month, Krafton followed most of ChatGPT’s recommendations,","relation":"supports","source_id":"s2"}],"assertion":"The court found that ChatGPT prepared a \"Response Strategy to a No-Deal Scenario\" recommending, among other steps, preemptive public framing, locking down publishing rights and access to the build pipeline, legal defense materials and team retention, and that Krafton followed most of these recommendations over the next month.","causal_attribution":"The opinion lists the strategy items and the steps Krafton took. The listed strategy does not include dismissing the executives. The dismissal decision is Krafton's."},{"id":"c4","status":"documented","evidence":[{"locator":"Krafton locked down Steam publishing rights to ensure Unknown Worlds could not publish Subnautica 2.","relation":"supports","source_id":"s1"},{"locator":"This severed Unknown Worlds’ practical ability to launch Subnautica 2.","relation":"supports","source_id":"s1"},{"locator":"On June 12, Krafton posted a message on the Unknown Worlds and Subnautica websites.","relation":"supports","source_id":"s1"},{"locator":"the Unknown Worlds team were shocked, since they had “nothing to do with” the message","relation":"supports","source_id":"s1"},{"locator":"On July 1, 2025, Krafton sent termination letters to","relation":"supports","source_id":"s1"},{"locator":"The letters identified a single ground for dismissal","relation":"supports","source_id":"s1"}],"assertion":"The court found that Krafton locked Unknown Worlds out of its Steam publishing rights, which the court found severed the studio's practical ability to launch its game, posted a message on the studio websites without the studio's involvement, and on 1 July 2025 sent termination letters citing a single ground.","causal_attribution":"Court findings of fact. The opinion links the lockdown to ChatGPT's recommendation and does not state that ChatGPT recommended the July terminations."},{"id":"c5","status":"documented","evidence":[{"locator":"Krafton’s newly manufactured justifications for the terminations are pretextual.","relation":"supports","source_id":"s1"},{"locator":"had secretly entered semi-retirement and that all three leaders had executed massive downloads of company data","relation":"supports","source_id":"s1"}],"assertion":"The court found Krafton's later justifications for the terminations (semi-retirement of two founders and data downloads) were pretextual.","causal_attribution":"Judicial finding after an expedited trial. Krafton's arguments are its litigation positions."},{"id":"c6","status":"documented","evidence":[{"locator":"is hereby reinstated as CEO of Unknown Worlds","relation":"supports","source_id":"s1"},{"locator":"authority over the early access launch of Subnautica 2","relation":"supports","source_id":"s1"},{"locator":"the base earnout Testing Period is equitably extended by 258 days","relation":"supports","source_id":"s1"}],"assertion":"The court reinstated the CEO of Unknown Worlds, enjoined Krafton from impeding the studio CEO's authority over the game's early access launch, and extended the base earnout Testing Period by 258 days.","causal_attribution":"Specific performance remedy. The court declined to return the two founders to their earlier roles."},{"id":"c7","status":"documented","evidence":[{"locator":"had deleted specific, relevant ChatGPT logs.","relation":"supports","source_id":"s1"},{"locator":"This particular chat was deleted.","relation":"supports","source_id":"s1"}],"assertion":"The CEO admitted at trial to deleting relevant ChatGPT logs, including the particular chat in which the earnout question was put to the chatbot.","causal_attribution":"A limit on the record. It bears on what the chatbot said, not on the court's findings about the terminations."},{"id":"c8","status":"documented","evidence":[{"locator":"formed an internal task force, dubbed “Project X.”","relation":"supports","source_id":"s1"},{"locator":"The task force’s mandate was to either negotiate a “deal” on the earnout or execute a “Take Over” of Unknown Worlds.","relation":"supports","source_id":"s1"},{"locator":"formed an internal taskforce, dubbed Project X.","relation":"supports","source_id":"s2"},{"locator":"The task force’s mandate was to either negotiate a ‘deal’ on the earnout or execute a ‘takeover’ of Unknown Worlds,","relation":"supports","source_id":"s3"}],"assertion":"The court found that, at ChatGPT's suggestion, the Krafton CEO formed an internal task force, Project X, whose mandate was either to negotiate a deal on the earnout or to execute a takeover of Unknown Worlds.","causal_attribution":"Finding of fact in the opinion, relayed by the Guardian and IGN. The task force's mandate is Krafton's; the opinion attributes only the suggestion to form it to ChatGPT."},{"id":"c9","status":"documented","evidence":[{"locator":"began to explore options, including firing the Key Employees.","relation":"supports","source_id":"s1"}],"assertion":"The court found that the Krafton CEO had begun exploring options, including firing the Key Employees, before the 2 June 2025 warning and the ChatGPT consultation that followed it.","causal_attribution":"Contrary evidence on causation: the idea of dismissal preceded the chatbot consultation."},{"id":"c10","status":"documented","evidence":[{"locator":"It also suggested a “key summary of responses” Krafton could deliver to the Key Employees:","relation":"supports","source_id":"s1"},{"locator":"Krafton’s seizure of Steam and other systems forced","relation":"supports","source_id":"s1"},{"locator":"to the negotiating table, and they began to discuss a solution.","relation":"supports","source_id":"s1"},{"locator":"It might actually be easier to just do a takeover.","relation":"supports","source_id":"s1"},{"locator":"responded: “Set a date.”","relation":"supports","source_id":"s1"}],"assertion":"The court found that ChatGPT's strategy also suggested a key summary of responses Krafton could deliver to the Key Employees, that Krafton's seizure of Steam and other systems forced the studio's CEO to the negotiating table, and that when negotiations stalled Krafton's Head of Strategy wrote on 27 June 2025 that it might be easier to just do a takeover, to which the Krafton CEO replied to set a date.","causal_attribution":"Findings of fact from Slack messages and trial testimony. They connect the ChatGPT-recommended lockdown to the negotiations and the takeover decision; the takeover decision itself is the Krafton executives'."},{"id":"c11","status":"reported","evidence":[{"locator":"have reached a mutual settlement and agreed to dismiss all pending legal proceedings,","relation":"supports","source_id":"s3"},{"locator":"Krafton said in a regulatory filing Wednesday that the company and the lead developers behind the globally popular Subnautica video game franchise decided to drop their court fight over controversial earn out bonuses.","relation":"supports","source_id":"s4"},{"locator":"officially settled the dispute and decided to withdraw all pending lawsuits as of Tuesday, the filing read.","relation":"supports","source_id":"s4"}],"assertion":"Krafton said in a statement to IGN that Krafton, Unknown Worlds and the three executives had reached a mutual settlement and agreed to dismiss all pending legal proceedings, and Yonhap, via The Korea Times, reported a Krafton regulatory filing on Wednesday 1 July 2026 saying the dispute was settled and all pending lawsuits withdrawn as of Tuesday 30 June 2026.","causal_attribution":"Both outlets relay Krafton's own announcement (a statement to IGN and a regulatory filing), one origin. The settlement terms beyond dismissal are not published by Krafton."},{"id":"c12","status":"reported","evidence":[{"locator":"had stepped down, leaving the company to look for a new boss.","relation":"supports","source_id":"s3"},{"locator":"The entire staff of Subnautica 2 developer Unknown Worlds will get their bonuses after parent company Krafton reached a settlement with studio leadership.","relation":"supports","source_id":"s3"}],"assertion":"IGN, citing Bloomberg, reported that the studio's CEO stepped down as part of the settlement and that the studio's entire staff would receive their bonuses.","causal_attribution":"Second-hand report (IGN relaying Bloomberg); Krafton's statement does not confirm these terms."},{"id":"c13","status":"reported","evidence":[{"locator":"Krafton said in a statement that it disagreed with the ruling and was evaluating its options and remained focused on delivering the best possible game for fans.","relation":"supports","source_id":"s2"}],"assertion":"Krafton said in a statement after the ruling that it disagreed with the ruling and was evaluating its options.","causal_attribution":"Krafton's litigation position as reported by the Guardian."}],"effects":[{"label":"Three studio executives terminated without valid Cause and removed from operational control during a takeover campaign that followed a CEO's ChatGPT strategy","claim_id":"c1","direction":"negative"}],"sources":[{"id":"s1","url":"https://courts.delaware.gov/Opinions/Download.aspx?id=392880","kind":"court_record","access":"read","language":"en","translation_note":"","independence_group":"delaware-chancery-opinion-2026-03"},{"id":"s2","url":"https://www.theguardian.com/technology/2026/mar/18/subnautica-2-publisher-krafton-ceo-reinstated-ai-chatgpt-failed-bid-avoid-paying-bonus","kind":"journalistic","access":"read","language":"en","translation_note":"","independence_group":"delaware-chancery-opinion-2026-03"},{"id":"s3","url":"https://www.ign.com/articles/subnautica-2-developers-entire-staff-get-their-bonuses-after-krafton-agrees-to-settlement-but-ceo-steps-down","kind":"journalistic","access":"read","language":"en","translation_note":"","independence_group":"krafton-settlement-announcement-2026-07"},{"id":"s4","url":"https://www.koreatimes.co.kr/business/companies/20260701/krafton-unknown-worlds-settle-legal-dispute-over-bonuses","kind":"journalistic","access":"read","language":"en","translation_note":"","independence_group":"krafton-settlement-announcement-2026-07"}],"version":1,"ai_roles":["institutional_use"],"contexts":["work"],"unknowns":["The date of the CEO's first ChatGPT use is not given beyond the sequence in the opinion (a legal warning on 2 June 2025 followed by the ChatGPT consultation). The incident date is the month.","The content of the deleted ChatGPT chat is unknown.","Whether ChatGPT recommended or was asked about terminating the executives is not stated in the opinion; the opinion records that the Krafton CEO had explored firing them before consulting it.","Phase Two (damages and earnout impairment) was never decided: the parties settled in July 2026, and the settlement terms beyond dismissal of all proceedings are known only from IGN's account of a Bloomberg report.","Whether Krafton appealed the March 2026 opinion before the settlement is not stated in the inspected sources.","Where the executives lived and worked is not stated in the inspected passages; the Korea Times describes the studio as a U.S. game studio and the opinion as a California corporation, which does not locate the lockout or the terminations.","The end of the harmful series is not a single date: termination letters were sent on 1 July 2025 effective 31 July, and the court counted the studio CEO's ouster as lasting until 16 March 2026."],"geography":{"basis":"The opinion is from the Delaware Court of Chancery (court country US). It describes Krafton as a South Korean company and the studio as a California corporation, and the Korea Times calls the studio a U.S. game studio; none of the inspected sources states where the executives were when locked out and terminated, so event and affected-person countries are left unknown rather than inferred from incorporation or headquarters.","court_countries":["US"],"event_countries":[],"affected_person_countries":[]},"publication":{"basis":"The full Delaware Court of Chancery opinion was read for the ChatGPT passages, the termination chronology, the holding and the remedy; the Guardian's report of the ruling and the IGN and Korea Times/Yonhap reports of the July 2026 settlement were read in full. Claims drawn from the opinion are court findings and are documented; Krafton's statements and the settlement terms are reported and attributed. The AI output (the chatbot's strategy and task-force suggestion), Krafton's use of it (the Steam lockdown as ChatGPT recommended) and the consequence for the three executives (loss of operational control and termination found to be without valid Cause) are cited separately. The affected executives are described by role.","reviewed_on":"2026-10-10"},"ai_involvement":{"basis":"Output: the court opinion finds that Krafton's CEO consulted ChatGPT after being warned that a dismissal with cause would not eliminate the earnout, that the chatbot answered the earnout would be difficult to cancel, that at ChatGPT's suggestion the CEO formed the Project X task force whose mandate was a deal or a takeover of the studio, and that ChatGPT prepared a \"Response Strategy to a No-Deal Scenario\" recommending, among other steps, locking down Steam and console publishing rights and build-pipeline access and a key summary of responses to deliver to the Key Employees. Use: the court finds that over the next month Krafton followed most of ChatGPT's recommendations and that, as ChatGPT recommended, it locked down the studio's Steam publishing rights. Consequence: the court finds the lockdown severed the studio's practical ability to launch its game and forced its CEO to the negotiating table, that the takeover and the 1 July 2025 terminations followed when negotiations stalled, and holds that Krafton breached the agreement by improperly seizing operational control from the Key Employees and terminating them without valid Cause. Limits: the strategy the opinion quotes does not list dismissing the executives, the CEO had explored firing them before consulting ChatGPT, and the first chat log was deleted, so the chatbot's first replies are known only through Slack messages.","status":"supported"},"person_relations":["informed_decision_about"]},"name":"Delaware court finds Krafton fired three Unknown Worlds executives without valid cause in a takeover that followed its CEO's ChatGPT strategy","summary":"A Delaware Court of Chancery opinion of 16 March 2026 finds that in June 2025 the CEO of Krafton, after being warned that a dismissal with cause would not eliminate a contingent earnout obligation, consulted ChatGPT. The chatbot answered that the earnout would be difficult to cancel, suggested an internal task force (Project X) whose mandate was to negotiate a deal or execute a takeover of the studio, and prepared a \"Response Strategy to a No-Deal Scenario\" that recommended locking down the studio's Steam and console publishing rights and build-pipeline access. The court found that Krafton followed most of these recommendations over the next month: it locked the studio out of its Steam publishing rights, which severed the studio's practical ability to launch its game and forced the studio's CEO to the negotiating table, posted a message on the studio's websites, and on 1 July 2025 sent termination letters to the studio's two founders and its CEO. The court held that Krafton breached the acquisition agreement by terminating the three executives without valid Cause and by improperly seizing operational control, reinstated the studio's CEO and extended the earnout period by 258 days. The opinion records that the Krafton CEO had begun exploring options including firing the executives before consulting ChatGPT, and does not find that ChatGPT recommended the terminations. Krafton said it disagreed with the ruling. On 1 July 2026 Krafton announced a mutual settlement with the studio and the three executives dismissing all pending proceedings; IGN, citing Bloomberg, reported that the studio's CEO stepped down as part of it.","incidentDate":"2025-06-01","incidentKind":"bounded_series","incidentDatePrecision":"month","exposurePattern":"unknown","reportedDate":"2026-03-16","aiSystem":"ChatGPT (OpenAI)","aiProduct":"ChatGPT","aiCompany":"OpenAI","severity":"medium","verificationStatus":"credible","harmCategories":[],"harmOutcomes":["professional_harm"],"harmOutcomeSummary":"The court found three studio executives were terminated without valid Cause and had operational control taken from them in a campaign that followed most of a ChatGPT-prepared strategy; one was reinstated by court order in March 2026 and the dispute was settled in July 2026.","frameworkFacets":[],"causationStatus":"unclear","participantUsersAffectedMin":0,"otherPeopleHarmedMin":3,"affectedCountStatus":"exact","affectedCountEvidence":"The opinion finds three Key Employees (two founders and the CEO) were terminated without valid Cause. Former shareholders represented by the plaintiff, studio staff and players are not counted.","victimAgeRange":"adult","platformType":"assistant","outcomeStatus":"resolved","primarySourceUrl":"https://courts.delaware.gov/Opinions/Download.aspx?id=392880","primarySourceLabel":"Delaware Court of Chancery opinion, Fortis Advisors v. Krafton, C.A. No. 2025-0805-LWW, 16 March 2026","firstPublishedAt":"2026-09-29T13:28:25.863533+00:00","updatedAt":"2026-10-10T03:20:34.158615+00:00","scopeVersion":"facts-v3","tags":["historical-2025","informed-decision"]},{"id":"2025-ar-esquel-criminal-judgment-annulled-judge-generative-ai-draft","caseFacts":{"claims":[{"id":"c1","status":"reported","evidence":[{"locator":"decidió el miércoles anular de oficio una sentencia dictada el 4 de junio pasado","relation":"supports","source_id":"s1"},{"locator":"anular de oficio una sentencia porque el juez la delegó en un ChatGPT","relation":"supports","source_id":"s3"},{"locator":"La sentencia anulada había condenado al imputado a dos años y seis meses de prisión efectiva por robo simple","relation":"supports","source_id":"s2"}],"assertion":"On 4 June 2025 a criminal judge in Esquel convicted a defendant of simple robbery and sentenced the defendant to two years and six months of effective imprisonment, and the Penal Chamber of Esquel later annulled that judgment.","causal_attribution":"Sources give the annulment as the chamber's response to the judge's AI use."},{"id":"c2","status":"reported","evidence":[{"locator":"Aquí tienes el punto IV reeditado, sin citas y listo para copiar y pegar","relation":"supports","source_id":"s1"},{"locator":"el magistrado utilizó la IA generativa, ni más ni menos, que en un aspecto central de la resolución en crisis como es la resolución de las nulidades de dos pruebas esenciales en el proceso","relation":"supports","source_id":"s2"},{"locator":"el juez de primera instancia incluyó accidentalmente una frase que revela el uso de un asistente de Inteligencia Artificial (IA) generativa","relation":"supports","source_id":"s3"}],"assertion":"The judgment contained the phrase “Aquí tienes el punto IV reeditado, sin citas y listo para copiar y pegar”, which the chamber took to show generative-AI use, including on the rulings on the nullity of two essential pieces of evidence.","causal_attribution":"The chamber's finding, as quoted in the press, connects the phrase to AI use. The full ruling text was not read."},{"id":"c3","status":"reported","evidence":[{"locator":"anular tanto la sentencia como el juicio que la precedió","relation":"supports","source_id":"s1"},{"locator":"declaración de nulidad de la sentencia y del juicio que la precede, con todo lo que ello significa para los justiciables","relation":"supports","source_id":"s3"},{"locator":"ordenando que el proceso se realice nuevamente con la intervención de otro magistrado","relation":"supports","source_id":"s1"},{"locator":"el tribunal ordenó reenviar el caso a otro juez penal, prohibiendo al Ministerio Público Fiscal exhibir en el nuevo juicio los videos omitidos en el anterior","relation":"supports","source_id":"s2"},{"locator":"A raíz de esto deberá realizarse nuevamente el juicio, con otro juez","relation":"supports","source_id":"s3"}],"assertion":"The chamber annulled the judgment and the preceding trial and ordered a new trial before another judge, and barred the prosecution from showing at the new trial the videos omitted at the first.","causal_attribution":"The annulment is attributed by the chamber to the judge's undisclosed and unsupervised AI use, which the chamber said prevented control of the judge's reasoning."},{"id":"c4","status":"reported","evidence":[{"locator":"remitir las actuaciones al Superior Tribunal de Justicia de Chubut (STJ), que deberá investigar el alcance ético y disciplinario de la conducta","relation":"supports","source_id":"s1"},{"locator":"pudo haber incumplido los recaudos de confidencialidad exigidos por el Acuerdo Plenario N° 5435 del STJ, al incluir nombres de las partes, testigos y peritos en el texto procesado por el asistente","relation":"supports","source_id":"s3"}],"assertion":"The chamber referred the matter to the Chubut Superior Court of Justice to investigate the judge's conduct and said the judge may have breached that court's confidentiality rules by including the names of parties, witnesses and experts in the text processed by the assistant.","causal_attribution":"The confidentiality point is the chamber's stated possibility and was not established as fact."},{"id":"c5","status":"reported","evidence":[{"locator":"presumiblemente ChatGPT","relation":"supports","source_id":"s1"},{"locator":"el juez la delegó en un ChatGPT","relation":"supports","source_id":"s3"}],"assertion":"Press coverage identifies the assistant as ChatGPT (La Nación as presumed, Perfil without qualification).","causal_attribution":"The ruling as quoted refers to a generative-AI assistant. The product name comes from the press."}],"effects":[{"label":"Conviction and trial annulled and retrial ordered; the chamber cites serious consequences for the parties","claim_id":"c3","direction":"negative"}],"sources":[{"id":"s1","url":"https://www.lanacion.com.ar/sociedad/el-juez-de-esquel-que-uso-chat-para-redactar-una-sentencia-sera-investigado-nid17102025/","kind":"news_report","access":"read","language":"es","translation_note":"Full body read in the original Spanish (17 October 2025) by the reviewing agent without a human translator. Reports the appellate ruling; whether the journalist saw the ruling text is not stated.","independence_group":"esquel-camara-ruling"},{"id":"s2","url":"https://documento.errepar.com/actualidad/justicia-anulada-se-invalida-un-fallo-por-el-uso-indebido-de-la-ia-20251016143159365","kind":"news_report","access":"read","language":"es","translation_note":"Article body is embedded in the page data and was decoded into bodies/gh-4-body.txt (16 October 2025). Read in Spanish without a human translator. It quotes the ruling. The linked full text of the ruling (documento.errepar.com/jurisprudencia/...) is gated and was not read.","independence_group":"esquel-camara-ruling"},{"id":"s3","url":"https://www.perfil.com/noticias/nea/inteligencia-artificial-anulan-sentencia-porque-un-juez-uso-un-chatgpt-y-una-frase-lo-delato.phtml","kind":"news_report","access":"read","language":"es","translation_note":"Full body read in Spanish (17 October 2025) without a human translator. It relays a statement of the Ministerio Publico Fiscal of Chubut and quotes the ruling.","independence_group":"esquel-camara-ruling"}],"version":1,"ai_roles":["institutional_use"],"contexts":["justice"],"unknowns":["The full text of the ruling was not read (the Errepar copy is gated), so all findings are as reported by three outlets that share the same underlying ruling.","The share of the judgment written by the assistant is unknown. The chamber said the pasted text leaves a gap too wide to tell.","The specific AI product is not established. No chat log was seen.","The annulment date of 15 October 2025 is derived from La Nación's 'el miércoles' (article of 17 October 2025) and the metadata of a gated Errepar ruling page; no cited article states the date.","The outcome of the Superior Court review and the date of the new trial are not reported in the sources read.","The defendant's own view of the annulment and any effect of the delay on the defendant or on other parties to the case is not reported."],"geography":{"basis":"The court is the Penal Chamber of Esquel in Chubut province, and La Nación calls the ground for the annulment unprecedented in Argentine justice. The defendant's residence and where the judge used the assistant are not stated.","court_countries":["AR"],"event_countries":[],"affected_person_countries":[]},"publication":{"basis":"Three Spanish-language outlets report the same appellate ruling, which annulled a criminal judgment and trial after a pasted assistant phrase exposed undisclosed generative-AI use. The chamber's stated consequences for the parties (annulment and a repeated trial) are reported, and the effect on the defendant is not. The outlets share one reporting chain, so claims are attributed.","reviewed_on":"2026-09-29"},"ai_involvement":{"basis":"The appellate chamber, as quoted by three outlets, found a pasted chat phrase in the ruling and concluded that a generative-AI assistant was used. The chamber described delegation of the decision as probable. The product name ChatGPT is stated by one outlet as presumed and by another without qualification.","status":"supported"},"person_relations":["made_decision_about"]},"name":"Argentina: appellate chamber annuls a robbery judgment and trial after the trial judge left a generative-AI chat phrase in the ruling","summary":"On 4 June 2025 a criminal judge in Esquel, Chubut province, convicted a defendant of simple robbery and imposed two years and six months of effective imprisonment. On 15 October 2025 the Penal Chamber of Esquel annulled the judgment and the trial, ordering a new trial before another judge. The chamber found that the ruling contained a pasted phrase from a generative-AI assistant (“Aquí tienes el punto IV reeditado, sin citas y listo para copiar y pegar”), that the AI had been used on the rulings on the nullity of two essential pieces of evidence, and that the judge had not exercised human supervision or recorded the use. Perfil's headline describes the assistant as ChatGPT and La Nación's text as presumably ChatGPT; the quoted ruling and the prosecution office statement refer only to a generative-AI assistant. The chamber referred the judge's conduct to the Chubut Superior Court of Justice for ethical and disciplinary review.","incidentDate":"2025-06-04","incidentEndDate":"2025-10-15","incidentKind":"bounded_series","incidentDatePrecision":"day","exposurePattern":"single_interaction","reportedDate":"2025-10-16","aiSystem":"Generative AI assistant (reported as ChatGPT)","aiProduct":"ChatGPT (reported)","severity":"low","verificationStatus":"credible","harmCategories":[],"harmOutcomes":["legal_harm"],"harmOutcomeSummary":"Per press reports of the appellate ruling, the trial and conviction were annulled because of the judge's undisclosed generative-AI use and the case must be tried again before another judge. The chamber described serious consequences of this for the parties, the public and the State. The effect on the defendant is not reported and the annulment may favour the defendant.","frameworkFacets":[],"causationStatus":"supported","participantUsersAffectedMin":0,"otherPeopleHarmedMin":0,"affectedCountStatus":"unquantified","affectedCountEvidence":"The chamber refers to serious consequences of the annulment for 'los justiciables' (the parties to the proceeding), the public and the State without a count. No inspected source reports harm to the defendant specifically, and annulment of a conviction may be favourable to the defendant.","victimAgeRange":"unknown","platformType":"assistant","primarySourceUrl":"https://www.lanacion.com.ar/sociedad/el-juez-de-esquel-que-uso-chat-para-redactar-una-sentencia-sera-investigado-nid17102025/","primarySourceLabel":"La Nación, 'El juez de Esquel que usó Chat para redactar una sentencia será investigado', 17 October 2025","firstPublishedAt":"2026-09-29T12:42:03.091102+00:00","updatedAt":"2026-09-30T01:16:55.894793+00:00","scopeVersion":"facts-v3","tags":["historical-2025"]},{"id":"2025-nsw-resilient-homes-program-applicant-data-uploaded-to-chatgpt","caseFacts":{"claims":[{"id":"c1","status":"reported","evidence":[{"locator":"Some personal information provided by people during the RHP applications process was uploaded by a former temporary employee of the RA to the AI platform, ChatGPT.","relation":"supports","source_id":"s1"},{"locator":"The data shared was contained in a Microsoft Excel spreadsheet with 10 columns and more than 12,000 rows of information.","relation":"supports","source_id":"s1"},{"locator":"The upload took place between 12 and 15 March 2025.","relation":"supports","source_id":"s1"},{"locator":"This incident occurred when a former temporary employee uploaded information from a Microsoft Excel spreadsheet to an unauthorised third-party AI platform, ChatGPT.","relation":"supports","source_id":"s1"},{"locator":"Between 12 and 15 March 2025, personal information was uploaded by a former contractor of the RA to the Artificial Intelligence platform ChatGPT.","relation":"supports","source_id":"s4"},{"locator":"\"Personal information was uploaded by a former contractor of the RA to the artificial intelligence platform ChatGPT,\" an RA spokesperson said.","relation":"supports","source_id":"s2"},{"locator":"The authority said in a breach notice that the upload “was not authorised”.","relation":"supports","source_id":"s3"}],"assertion":"Between 12 and 15 March 2025 a former temporary staff member (called a former contractor in the first notice) of the NSW Reconstruction Authority uploaded a Microsoft Excel spreadsheet of Northern Rivers Resilient Homes Program information (10 columns, more than 12,000 rows) to ChatGPT; the Authority says the upload was not authorised and was not a cyber-attack.","causal_attribution":"The Authority's own notices report the upload, its dates and its unauthorised status; the ABC and iTnews relay those notices. The uploader's role is a former temporary staff member on the current page and a former contractor in the first notice and the press."},{"id":"c2","status":"reported","evidence":[{"locator":"2031 people who provided information to the RA for the RHP had some of their data uploaded to ChatGPT.","relation":"supports","source_id":"s1"},{"locator":"Name and contact details","relation":"supports","source_id":"s1"},{"locator":"Residential/mailing address","relation":"supports","source_id":"s1"},{"locator":"Date of birth","relation":"supports","source_id":"s1"},{"locator":"Sensitive personal information","relation":"supports","source_id":"s1"},{"locator":"Sensitive health information","relation":"supports","source_id":"s1"},{"locator":"Limited financial commentary, but not banking or financial details","relation":"supports","source_id":"s1"},{"locator":"no driver's licence numbers, Medicare numbers, passport numbers, or Tax File Numbers were disclosed in the breach.","relation":"supports","source_id":"s1"},{"locator":"Up to 3000 individuals may be impacted by the breach.","relation":"supports","source_id":"s4"},{"locator":"The breach shared the names and addresses of program applicants, as well as email addresses, phone numbers, and other personal and health information.","relation":"supports","source_id":"s2"}],"assertion":"The Authority says external forensic analysis confirmed that 2031 people who provided information for the program had some of their data uploaded to ChatGPT, including name and contact details, residential or mailing address, date of birth, sensitive personal and health information and limited financial commentary, and that no driver licence, Medicare, passport or Tax File Numbers were disclosed; its first notice had estimated up to 3000 people.","causal_attribution":"The Authority's statements report the count and data categories from its forensic analysis; the ABC relays the first notice's categories."},{"id":"c3","status":"reported","evidence":[{"locator":"There is no evidence that any of the uploaded data is publicly available online or has been accessed by a third party at this stage.","relation":"supports","source_id":"s1"},{"locator":"We believe the risk of misuse is low","relation":"supports","source_id":"s1"},{"locator":"There is no evidence that any information has been made public, however this cannot be ruled out","relation":"supports","source_id":"s4"}],"assertion":"The Authority says there is no evidence that the uploaded data is publicly available online or has been accessed by a third party; its first notice said this could not be ruled out, and it assesses the risk of misuse as low.","causal_attribution":"The Authority's own assessment. Whether OpenAI retained, deleted or used the uploaded data is not addressed in the inspected sources."},{"id":"c4","status":"reported","evidence":[{"locator":"we are deeply sorry for the distress it may cause for those involved in the program.","relation":"supports","source_id":"s1"},{"locator":"I feel concerned — the information with the RA was deeply personal,","relation":"supports","source_id":"s2"}],"assertion":"The Authority apologised for the distress the breach may cause, and a program participant whose flooded home was bought back told the ABC the information held by the Authority was deeply personal and that they felt concerned.","causal_attribution":"The concern is one participant's stated reaction; the ABC does not say whether this person is among the 2031. No identity misuse or other realised loss is reported."},{"id":"c5","status":"reported","evidence":[{"locator":"We are working with Social Futures to reach out to people who have been impacted and ID Support NSW, a government identity and cyber security support service, to assist anyone whose data may have been compromised.","relation":"supports","source_id":"s1"},{"locator":"The NSW Reconstruction Authority will provide compensation for any reasonable out of pocket expenses if any compromised identity documents need to be replaced.","relation":"supports","source_id":"s4"}],"assertion":"The Authority contacted affected people with ID Support NSW and Social Futures to offer support, and said it would compensate reasonable out-of-pocket expenses if compromised identity documents needed replacing.","causal_attribution":"The Authority's statements about its response."},{"id":"c6","status":"reported","evidence":[{"locator":"We are working with Cyber Security NSW to monitor the internet and dark web to see if any of this information is accessible online. The NSW Privacy Commissioner has also been notified.","relation":"supports","source_id":"s1"},{"locator":"The RA has also implemented controls to block the upload of personal information into AI tools.","relation":"supports","source_id":"s1"},{"locator":"An independent review into the data breach has now been completed, identifying several areas for improvement in RA’s governance, training and data security practices.","relation":"supports","source_id":"s1"}],"assertion":"The Authority notified the NSW Privacy Commissioner, had Cyber Security NSW monitor the internet and dark web, implemented controls to block the upload of personal information into AI tools, and completed an independent review that identified areas for improvement in its governance, training and data security practices.","causal_attribution":"The Authority's own account of its response; the review report itself was not inspected."},{"id":"c7","status":"reported","evidence":[{"locator":"The AI-related data breach was disclosed on what is a public holiday Monday in NSW, although it actually occurred over six months ago, between March 12 and 15.","relation":"supports","source_id":"s3"},{"locator":"said she received a preliminary report about the breach in July, and the final report on Friday.","relation":"supports","source_id":"s2"},{"locator":"\"I would prefer [for the RA] to advise people sooner, and I'm also informed they couldn't identify everybody, but I need to see the report first,\" she said.","relation":"supports","source_id":"s2"}],"assertion":"The breach was disclosed publicly on 6 October 2025, more than six months after the upload; the NSW Minister for Recovery told the ABC she had received a preliminary report about the breach in July and would have preferred people to be advised sooner.","causal_attribution":"iTnews's own observation of the disclosure timing and the Minister's statement to the ABC; the Authority attributes the delay to the time needed to identify affected people."}],"effects":[{"label":"personal and sensitive information of 2031 Resilient Homes Program participants was uploaded to ChatGPT without authorisation","claim_id":"c2","direction":"negative"},{"label":"a program participant reports concern about the deeply personal information held; the Authority apologised for possible distress","claim_id":"c4","direction":"negative"}],"sources":[{"id":"s1","url":"https://www.nsw.gov.au/departments-and-agencies/nsw-reconstruction-authority/resilient-homes-program-data-breach","kind":"official_statement","access":"read","language":"en","translation_note":"English original; full body read by curl on 2026-10-11 (page last updated 26 March 2026, with the 9 December 2025 update and FAQ on the same page).","independence_group":"nsw-ra-statement"},{"id":"s2","url":"https://www.abc.net.au/news/2025-10-06/data-breach-northern-rivers-resilient-homes-program-chatgpt/105855284","kind":"news_report","access":"read","language":"en","translation_note":"English original; full body read by curl on 2026-10-11. Relays the Authority statement; the participant and the Minister were interviewed separately within the same article.","independence_group":"nsw-ra-statement"},{"id":"s3","url":"https://www.itnews.com.au/news/nsw-gov-contractor-uploaded-excel-spreadsheet-of-flood-victims-data-to-chatgpt-620815","kind":"news_report","access":"read","language":"en","translation_note":"English original; full body read by curl on 2026-10-11. Relays the Authority breach notice; the disclosure-timing observation is the outlet's own.","independence_group":"nsw-ra-statement"},{"id":"s4","url":"https://www.nsw.gov.au/departments-and-agencies/nsw-reconstruction-authority/media-releases/northern-rivers-resilient-homes-program-data-breach","kind":"official_statement","access":"read","language":"en","translation_note":"English original; full body read by curl on 2026-10-11. Authority departmental release of the first notice, 6 October 2025.","independence_group":"nsw-ra-statement"}],"version":1,"ai_roles":["institutional_use"],"contexts":["privacy","public_services"],"unknowns":["Why the staff member uploaded the spreadsheet and what, if anything, was asked of ChatGPT; no ChatGPT output is described.","Whether OpenAI retained, deleted or used the uploaded data for training; the sources establish the upload to the service only.","Whether any participant suffered identity misuse or other realised loss; none is reported.","Whether the uploader was a temporary staff member or a contractor (the Authority uses both terms), and how many separate uploads occurred between 12 and 15 March 2025.","The findings of the independent review beyond the Authority's summary of it."],"geography":{"basis":"The data was held by the NSW Reconstruction Authority, a New South Wales government agency, and was uploaded by a former temporary staff member of that agency; the Authority describes the incident as a breach of its program data. The ABC describes the affected people as northern NSW residents of the Northern Rivers region. No court is involved.","court_countries":[],"event_countries":["AU"],"affected_person_countries":["AU"]},"publication":{"basis":"The Authority's notices, read in full on 2026-10-11, state that a staff member uploaded program participants' personal and sensitive information to ChatGPT without authorisation and that forensic analysis confirmed 2031 affected people. The disclosure of their information to a generative AI service without their knowledge is the recorded privacy harm. The upload rests on the Authority's own account, which the press relayed; no retention, training, public exposure or misuse is established. The uploader and affected people are not named.","reviewed_on":"2026-10-11"},"ai_involvement":{"basis":"The Authority states that a former temporary staff member uploaded a spreadsheet of program information to ChatGPT between 12 and 15 March 2025, that the upload was not authorised, and that forensic analysis confirmed 2031 people had some of their data uploaded. The reported transfer of their personal and sensitive information into the generative AI service is the AI connection to the privacy harm. This establishes the upload only. It does not establish retention by OpenAI, use for training, public exposure or a finding of illegality, and no ChatGPT output affecting participants is described.","status":"supported"},"person_relations":["received_personal_information"]},"name":"NSW Reconstruction Authority says a former temporary staff member uploaded 2031 Resilient Homes participants' personal data to ChatGPT without authorisation","summary":"The NSW Reconstruction Authority says that between 12 and 15 March 2025 a former temporary staff member (called a former contractor in its first notice) uploaded a Microsoft Excel spreadsheet of Northern Rivers Resilient Homes Program information, 10 columns and more than 12,000 rows, to ChatGPT, an AI platform the Authority had not authorised. The Authority disclosed the breach on 6 October 2025, estimating that up to 3000 people were affected, and later said external forensic analysis confirmed that 2031 people had some of their data uploaded, including names, contact details, addresses, dates of birth, sensitive personal and health information and limited financial commentary. It says no driver licence, Medicare, passport or Tax File Numbers were disclosed and that there is no evidence the data is publicly available or was accessed by a third party; whether OpenAI retained or used the data is not addressed. The Authority apologised, contacted affected people with ID Support NSW, offered to compensate the replacement of compromised identity documents and notified the NSW Privacy Commissioner. A program participant told the ABC the information was deeply personal and that they felt concerned. No misuse of the data is reported.","incidentDate":"2025-03-12","incidentEndDate":"2025-03-15","incidentKind":"bounded_series","incidentDatePrecision":"range","exposurePattern":"unknown","reportedDate":"2025-10-06","aiSystem":"ChatGPT (OpenAI), used through an unauthorised upload of program data by a person working for the NSW Reconstruction Authority","aiProduct":"ChatGPT","aiCompany":"OpenAI","severity":"low","verificationStatus":"credible","harmCategories":[],"harmOutcomes":["other_material_harm"],"harmOutcomeSummary":"The NSW Reconstruction Authority reports that personal and sensitive information of 2031 Resilient Homes Program participants was uploaded to ChatGPT without authorisation, a loss of privacy for those people; it reports no evidence of public exposure or third-party access, and no misuse is reported (the Authority's notices and press coverage relaying them).","frameworkFacets":[],"causationStatus":"alleged","participantUsersAffectedMin":0,"otherPeopleHarmedMin":2031,"affectedCountStatus":"exact","affectedCountEvidence":"The Authority states that external forensic analysis confirmed 2031 people had some of their data uploaded to ChatGPT. The upload of their personal information to the AI service is itself the privacy harm recorded, so those 2031 people are counted as other people harmed. The first notice's estimate of up to 3000 is not used; the 12,000 rows are rows, not people. The participant quoted by the ABC is not separately counted because the ABC does not say whether that person is among the 2031. The uploader is not counted.","victimAgeRange":"unknown","jurisdiction":"AU-NSW","platformType":"chatbot","outcomeType":"investigation_opened","outcomeStatus":"resolved","primarySourceUrl":"https://www.nsw.gov.au/departments-and-agencies/nsw-reconstruction-authority/resilient-homes-program-data-breach","primarySourceLabel":"NSW Reconstruction Authority, updated 26 March 2026: Resilient Homes Program data breach","firstPublishedAt":"2026-09-29T12:41:19.196147+00:00","updatedAt":"2026-10-11T03:10:38.11789+00:00","scopeVersion":"facts-v3","tags":["historical-2025"]},{"id":"2025-stephens-county-oklahoma-judge-chatgpt-fake-citations-paternity-order","caseFacts":{"claims":[{"id":"c1","status":"reported","evidence":[{"locator":"'Wheeler denied a request from one of the parents' attorneys to have the other parent undergo a psychological test—and in his order denying the request, he pointed to multiple previous Oklahoma cases he said backed him up.'","relation":"supports","source_id":"s2"},{"locator":"'The 2025 order in a child paternity case'; 'He came under scrutiny after he reprimanded attorney [the attorney] last November in an order in the paternity case for asking for a psychological evaluation of the child’s father.'","relation":"supports","source_id":"s1"},{"locator":"'The issue came to light after a request for a mental evaluation was denied and a subsequent court order was filed.'","relation":"supports","source_id":"s3"}],"assertion":"In November 2025, in a child paternity and custody case, Judge Wheeler issued an order denying a request by one parent's attorney for a psychological evaluation of the other parent, citing Oklahoma cases he said supported the decision.","causal_attribution":"Court history as reported by The Oklahoman and KFOR and described by the mother; the order itself was not read."},{"id":"c2","status":"reported","evidence":[{"locator":"'The judge told the OSBI he only used ChatGPT for research and created the order himself, DA Jason Hicks wrote. The judge, however, also “indicated that at least two of the case citations included in his order and produced by ChatGPT do not exist.”'","relation":"supports","source_id":"s1"},{"locator":"'Stephens County District Attorney Jason Hicks says Wheeler admitted to OSBI agents he cited \"at least two\" cases ChatGPT fed him that \"do not exist.\"'","relation":"supports","source_id":"s2"},{"locator":"'Hicks wrote that Wheeler acknowledged using case citations generated by ChatGPT and that at least two citations included in the order did not exist.'","relation":"supports","source_id":"s3"},{"locator":"'Associate District Judge Lawrence Wheeler used OpenAI's ChatGPT for research while drafting the order, which cited two cases that do not exist, according to state records.'","relation":"supports","source_id":"s4"}],"assertion":"According to the district attorney's 17 August 2026 letter, the judge told the OSBI that he used ChatGPT for research and wrote the order himself, and indicated that at least two case citations in the order, produced by ChatGPT, do not exist.","causal_attribution":"The judge's own statement to investigators as relayed in the district attorney's letter; all four reports rely on that letter, which was not read. The judge's attorney declined to comment."},{"id":"c3","status":"reported","evidence":[{"locator":"'The judge issued the reprimand “for stooping to such frivolous trial tactics” and cautioned that future similar filings may be subject to further sanctions.'; '[The attorney] challenged the order at the Oklahoma Supreme Court in February. He told justices that Wheeler based his order on two cases, Cummings v. Cimarron Elevator Co. and Hawkins v. Linhart, that do not exist.'; '[The attorney] dismissed his challenge in March after Wheeler vacated the order. Wheeler is no longer involved in the paternity case.'","relation":"supports","source_id":"s1"},{"locator":"'She said her attorney at the time challenged the ruling and began reviewing the legal authorities cited in the order.'","relation":"supports","source_id":"s3"}],"assertion":"The order reprimanded the attorney who had asked for a psychological evaluation of the child's father 'for stooping to such frivolous trial tactics' and warned of further sanctions; that attorney challenged it at the Oklahoma Supreme Court in February 2026, saying it relied on two nonexistent cases, and dismissed the challenge in March after Wheeler vacated the order. The mother told News 9 that her attorney at the time challenged the ruling.","causal_attribution":"The Oklahoman's account of the court filings and the mother's account to News 9; the Supreme Court filings were not read."},{"id":"c4","status":"reported","evidence":[{"locator":"'[The mother] said learning the citations were allegedly fabricated was alarming because the order involved the custody of a child.'; '“Then it became very concerning to me that somebody would do something like that over a child custody case,” she said.'; '“He then turned around and decided to recuse himself from my case,” she said. “He called my attorney that I had at the time and apologized to him.”'","relation":"supports","source_id":"s3"}],"assertion":"The mother told News 9 that learning the citations were allegedly fabricated was alarming because the order involved the custody of a child, and that the judge later recused himself from her case and apologised to her attorney.","causal_attribution":"The mother's first-person account in a broadcast interview."},{"id":"c5","status":"reported","evidence":[{"locator":"'The OSBI investigated the judge at the request of the Oklahoma Council on Judicial Complaints.'; 'The OSBI also looked into allegations “that Judge Wheeler offered to amend the order in exchange for a dismissal of a writ seeking relief from the Oklahoma Supreme Court” and that he “eavesdropped on jury deliberations,” the DA disclosed in the letter.'; 'Prosecutors in the AG’s office “thoroughly reviewed the results of the investigation and determined the evidence does not support a criminal prosecution,” the AG’s press secretary, Leslie Berger, said on Sept. 2.'","relation":"supports","source_id":"s1"},{"locator":"'The Criminal Justice Division of the Attorney General’s Office thoroughly reviewed the results of the investigation and determined the evidence does not support a criminal prosecution. The Oklahoma Supreme Court and the Court on the Judiciary possess ultimate jurisdiction over disciplinary actions for allegations of inappropriate behavior by judges regardless of whether the behavior constitutes a violation of a criminal statute.'","relation":"supports","source_id":"s3"}],"assertion":"The OSBI investigated the judge at the request of the Council on Judicial Complaints, including allegations that he offered to amend the order in exchange for dismissal of the writ and that he eavesdropped on jury deliberations; the attorney general's office said the evidence did not support criminal prosecution and that judicial discipline rests with the Oklahoma Supreme Court and the Court on the Judiciary.","causal_attribution":"District attorney's letter and attorney general's statement as reported; no disciplinary finding has been made."}],"effects":[{"label":"the mother faced an order denying a psychological-evaluation request, with the judge citing authorities reported to have been generated by ChatGPT","claim_id":"c1","direction":"negative"},{"label":"the attorney who requested the evaluation was reprimanded in the order and sought relief from the Oklahoma Supreme Court before the order was vacated","claim_id":"c3","direction":"negative"},{"label":"the mother describes learning of the fabricated citations in her child custody case as alarming","claim_id":"c4","direction":"negative"}],"sources":[{"id":"s1","url":"https://minnlawyer.com/2026/09/09/oklahoma-judge-ai-fabricated-case-citations/","kind":"news_report","access":"read","language":"en","translation_note":"The Oklahoman's report (USA TODAY Network via Reuters Connect) as republished by Minnesota Lawyer; read in full from the Wayback capture of 10 September 2026 because the live page returned 403. Its ChatGPT-admission passages relay the district attorney's letter; its account of the reprimand, the Supreme Court challenge and the vacatur draws on court filings and the attorney. Reinspected in the retained complete English body on 2026-10-06. All sources share the district-attorney-letter chain for AI involvement, so grouped conservatively together; the News 9 maternal interview is separate testimony about her experience, not a second verification of AI use.","independence_group":"wheeler-da-letter"},{"id":"s2","url":"https://www.aol.com/articles/oklahoma-judge-admitted-citing-fake-030000000.html","kind":"news_report","access":"read","language":"en","translation_note":"KFOR (Oklahoma City) report syndicated on AOL, read in full on 2026-09-29. Its account of the ChatGPT admission relays the same district attorney letter. Reinspected in the retained complete English body on 2026-10-06. All sources share the district-attorney-letter chain for AI involvement, so grouped conservatively together; the News 9 maternal interview is separate testimony about her experience, not a second verification of AI use.","independence_group":"wheeler-da-letter"},{"id":"s3","url":"https://www.news9.com/oklahoma-city-news/stephens-county-mother-speaks-out-after-judge-accused-of-using-chatgpt-in-custody-case","kind":"news_report","access":"read","language":"en","translation_note":"News 9 (Griffin Media) interview with the mother, read in full on 2026-09-29. Its description of the ChatGPT admission also relays the district attorney's letter; the mother's own account is independent of it. Reinspected in the retained complete English body on 2026-10-06. All sources share the district-attorney-letter chain for AI involvement, so grouped conservatively together; the News 9 maternal interview is separate testimony about her experience, not a second verification of AI use.","independence_group":"wheeler-da-letter"},{"id":"s4","url":"https://hoodline.com/2026/09/stephens-county-judge-faces-discipline-over-chatgpt-written-court-order/","kind":"news_report","access":"read","language":"en","translation_note":"Hoodline rewrite of the Reuters report, read in full on 2026-09-29. Reuters relays the same district attorney letter. Reinspected in the retained complete English body on 2026-10-06. All sources share the district-attorney-letter chain for AI involvement, so grouped conservatively together; the News 9 maternal interview is separate testimony about her experience, not a second verification of AI use.","independence_group":"wheeler-da-letter"}],"version":1,"ai_roles":["institutional_use"],"contexts":["justice","relationships"],"unknowns":["The text of the order and whether the nonexistent citations were decisive for denying the evaluation and for the reprimand.","Whether any disciplinary proceeding follows from the Council on Judicial Complaints referral.","The current state of the custody case after the vacatur and the judge's departure from it.","The exact day in November 2025 on which the order was issued."],"geography":{"basis":"The Oklahoman, KFOR and News 9 locate the order and proceedings in Stephens County, Oklahoma, and the Oklahoma Supreme Court. These establish the event and court countries as US. News 9 calls the interviewee a Stephens County mother but gives no explicit residence evidence for either counted person, so affected_person_countries is unknown.","court_countries":["US"],"event_countries":["US"],"affected_person_countries":[]},"publication":{"basis":"This reported event connects specific ChatGPT-generated citations to an adverse order issued by a human judge. The Oklahoman and KFOR describe their use in the order, The Oklahoman reports the attorney's reprimand and successful challenge, and News 9 carries the mother's account. The AI admission is reported through one district attorney's letter, not independently verified model output. The family and attorney remain unnamed. No broader custody consequence or disciplinary finding is inferred.","reviewed_on":"2026-10-06"},"ai_involvement":{"basis":"The Oklahoman, KFOR and News 9 relay a district attorney's letter saying the judge acknowledged using ChatGPT for research and incorporating at least two nonexistent citations it produced into his order. The cited authorities were presented in support of the human judge's denial of a requested evaluation and his reprimand of the attorney. This supports informed_decision_about for the affected mother and attorney. It does not establish that ChatGPT itself decided custody or communicated with either of them. The underlying letter and order were not inspected, and whether the citations determined the outcome is unknown.","status":"supported"},"person_relations":["informed_decision_about"]},"name":"Prosecutor: judge used ChatGPT citations in Oklahoma custody order later vacated","summary":"In November 2025 Stephens County Associate District Judge Lawrence Wheeler issued an order in a child paternity and custody case that denied one parent's request for a psychological evaluation of the other parent and reprimanded the requesting parent's attorney 'for stooping to such frivolous trial tactics'. The attorney challenged the order at the Oklahoma Supreme Court in February 2026, telling the justices that it relied on two cases that do not exist; the challenge was dismissed in March after Wheeler vacated the order, and he is no longer on the case. According to a 17 August 2026 letter from the Stephens County district attorney, Wheeler told the Oklahoma State Bureau of Investigation that he used ChatGPT for research and wrote the order himself, and that at least two citations in it produced by ChatGPT do not exist. The mother in the case told News 9 that learning the order's citations were allegedly fabricated was alarming because it concerned the custody of a child. The attorney general's office declined criminal prosecution; judicial discipline remains possible.","incidentDate":"2025-11-01","incidentKind":"single_event","incidentDatePrecision":"month","exposurePattern":"unknown","reportedDate":"2026-09-09","aiSystem":"ChatGPT (OpenAI), used by the judge for legal research while preparing the order","aiProduct":"ChatGPT","aiCompany":"OpenAI","severity":"low","verificationStatus":"credible","harmCategories":[],"harmOutcomes":["legal_harm","professional_harm"],"harmOutcomeSummary":"According to the district attorney's letter as reported by The Oklahoman and KFOR, a judge acknowledged that his custody-case order contained at least two nonexistent citations supplied by ChatGPT. The order denied a request for a psychological evaluation and reprimanded the requesting attorney. The attorney challenged it at the Oklahoma Supreme Court before it was vacated. The mother described the discovery as alarming in a News 9 interview. The order and letter have not been inspected directly.","frameworkFacets":[],"causationStatus":"supported","participantUsersAffectedMin":0,"otherPeopleHarmedMin":2,"affectedCountStatus":"documented_minimum","affectedCountEvidence":"Two people affected through the judge's use of ChatGPT: the mother, whose request for a psychological evaluation was denied by the order and who describes the discovery as alarming (News 9), and the attorney who requested the evaluation and was reprimanded in the order and challenged it at the Oklahoma Supreme Court (The Oklahoman). The child and the other parent are not counted because no harm to them is reported. The judge, the AI user, is not counted: he faces a possible disciplinary process but no sanction has been reported and the complaint bundled other allegations.","victimAgeRange":"adult","jurisdiction":"US-OK","platformType":"assistant","outcomeType":"investigation_opened","outcomeStatus":"ongoing","primarySourceUrl":"https://minnlawyer.com/2026/09/09/oklahoma-judge-ai-fabricated-case-citations/","primarySourceLabel":"The Oklahoman via Minnesota Lawyer, 9 September 2026: judge used ChatGPT citations in custody order","firstPublishedAt":"2026-09-29T09:10:18.787549+00:00","updatedAt":"2026-10-06T04:48:44.069668+00:00","scopeVersion":"facts-v3","tags":["justice","family-court","judge","fabricated-citations","chatgpt","institutional-use","oklahoma"]}]}