{"meta":{"exportedAt":"2026-10-23T08:49:16.612Z","formatVersion":2,"selection":{"q":"eu","system":"","harm":"","context":"","country":"","role":"","relation":"unknown","evidence":"","year":"","response":"regulatory","severity":"","verification":"","view":"incidents","sort":"added"},"totalIncidents":4,"coverage":{"cases":4,"countries":3,"languages":3,"unknownLocation":0,"locationPending":0,"unknownLanguage":0,"unknownDate":0,"lawsuits":0,"regulatory":4,"minors":1,"coreRelations":0,"contextualRelations":0,"mixedRelations":0,"unknownRelations":4,"relationPending":1,"relationUnknown":3},"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":"2026-connecticut-court-revokes-self-represented-plaintiffs-e-filing-access-over-hidden-instructions-aimed-at-ai","caseFacts":{"claims":[{"id":"c1","status":"reported","evidence":[{"locator":"contains hidden text set in tiny-point type and colored white, so that it is effectively invisible to a person reading the document; In substance it directs that any artificial-intelligence model reviewing or receiving the document ensure that its output agrees with the plaintiff's filing; also filed July 24, 2026, contains a further, abbreviated, instance of the concealed, tiny-point white text","relation":"supports","source_id":"s1"},{"locator":"The offending text directed any AI system reviewing the document to ensure textual outputs agreed with the plaintiff’s arguments","relation":"supports","source_id":"s2"}],"assertion":"A motion for default filed by the self-represented plaintiff on 24 July 2026 contained hidden text in tiny white type directing any AI model reviewing it to ensure its output agreed with the filing and to work toward reversing the Chief Clerk's denial of his default request; a notice filed the same day carried an abbreviated version.","causal_attribution":"The court's findings as excerpted by Volokh; Ars Technica reports the same decision."},{"id":"c2","status":"reported","evidence":[{"locator":"The Court sent its Notice of [a] Hearing [to consider whether there was a sanctionable prompt injection] on July 31, 2026.; On the morning of the Hearing, the plaintiff again hid text in pleadings.; as a dutiful citizen auditing the Court's Al systems; The plaintiff replied that he did so as a joke","relation":"supports","source_id":"s1"}],"assertion":"After the court sent notice on 31 July 2026 of a hearing on the hidden text, the plaintiff hid further messages in later filings, including a nonsense message, a short greeting and a link to a video; he told the court the first instruction was an audit of the court's AI systems and the later messages were jokes.","causal_attribution":"The court's account of the hearing, as excerpted."},{"id":"c3","status":"reported","evidence":[{"locator":"The technique has a name, \"prompt injection.\"; The Court does not find the plaintiff credible that he only added the prompt to \"audit\" the Court's use of artificial-intelligence systems.; The plaintiff's ability to file matters electronically through the Court's e-filing system is rescinded. Any future pleadings or exhibits by the plaintiff shall be filed in person, on paper, at the clerk's office.","relation":"supports","source_id":"s1"},{"locator":"declined to order monetary penalties","relation":"supports","source_id":"s2"}],"assertion":"The court found the instruction a prompt injection, did not credit the audit explanation, and rescinded the plaintiff's ability to file electronically, requiring future pleadings to be filed in person on paper at the clerk's office; Ars Technica reports that no monetary penalty was ordered.","causal_attribution":"The court's order as excerpted; the absence of a monetary penalty is Ars Technica's report."},{"id":"c4","status":"reported","evidence":[{"locator":"The Connecticut Judicial Branch does not utilize an artificial-intelligence system to review or decide filings, although a number of court systems elsewhere do so. The undersigned denied Docket Entry #177.00 on its merits working off a printed version of the motion, so the hidden instruction had no impact on a ruling.","relation":"supports","source_id":"s1"},{"locator":"confirmed that the hidden text had no impact in a case where a man alleged a healthcare provider was improperly withholding access to records","relation":"supports","source_id":"s2"}],"assertion":"The court stated that the Connecticut Judicial Branch does not use an AI system to review or decide filings and that the judge denied the motion on its merits from a printed version, so the hidden instruction had no impact on a ruling.","causal_attribution":"The court's statement, as excerpted and as reported by Ars Technica."},{"id":"c5","status":"reported","evidence":[{"locator":"told Reuters yesterday that he maintains that his intent was to audit the court","relation":"supports","source_id":"s2"}],"assertion":"The plaintiff told Reuters that he maintains his intent was to audit the court, according to Ars Technica.","causal_attribution":"The plaintiff's own account, relayed by Ars Technica; the court did not find it credible."}],"effects":[{"label":"e-filing access rescinded; future filings must be made on paper at the clerk's office","claim_id":"c3","direction":"negative"}],"sources":[{"id":"s1","url":"https://reason.com/volokh/2026/08/13/court-faults-self-represented-plaintiff-for-including-hidden-prompt-injection-in-court-filing/","kind":"legal_commentary","access":"read","language":"en","translation_note":"Read in full in English by the research agent (an AI) on 2026-10-18 (HTTP 200); no translation was needed. The post quotes long passages of the decision with ellipses.","independence_group":"connecticut-superior-court-decision-aug-2026"},{"id":"s2","url":"https://arstechnica.com/tech-policy/2026/08/suspecting-court-of-using-ai-man-injected-prompts-in-filings-to-try-to-win-case/","kind":"news","access":"read","language":"en","translation_note":"Read in full in English by the research agent (an AI) on 2026-10-18 (HTTP 200); no translation was needed.","independence_group":"connecticut-superior-court-decision-aug-2026"}],"version":1,"ai_roles":["unknown"],"contexts":["justice"],"unknowns":["The exact date of the memorandum of decision; the sources say it was issued in the week before 13 August 2026.","Whether the plaintiff used an AI tool to draft his filings or the hidden text; Ars Technica's suggestion that he was convinced by an AI system is not established by the decision excerpt read.","Whether the sanction was challenged and how the underlying records dispute ended."],"geography":{"basis":"The filings were made in, and the sanction issued by, the Connecticut Superior Court. The plaintiff's residence is not used.","court_countries":["US"],"event_countries":["US"],"affected_person_countries":[]},"publication":{"basis":"Published as a consequential-response case: the Connecticut Superior Court's decision, read through Volokh's excerpt and Ars Technica's report, rescinds a self-represented plaintiff's e-filing access because his filings carried hidden instructions aimed at AI systems. The court says no AI system reviews its filings and the text affected no ruling. The plaintiff is a private person and is not named.","reviewed_on":"2026-10-18"},"ai_involvement":{"basis":"The harm is the court's response to concealed instructions aimed at AI systems: the decision, as excerpted, records hidden white text in the plaintiff's 24 July 2026 filings directing any AI model reviewing them to make its output agree with the filing, treats this as a prompt injection and rescinds his e-filing access for it. The court states that the Connecticut Judicial Branch uses no AI system to review or decide filings and that the hidden text affected no ruling, so no relation describing an AI system's own action toward the plaintiff is established (unknown; consequential response).","status":"reported"},"person_relations":["unknown"]},"name":"Connecticut court revokes a self-represented plaintiff's e-filing access after finding hidden instructions aimed at AI systems in his filings","summary":"In a memorandum of decision issued in early August 2026, a Connecticut Superior Court judge found that a motion for default and a notice filed on 24 July 2026 by a self-represented plaintiff, in a suit against a healthcare provider over access to records, carried text in tiny white type that was invisible to a human reader but legible to software. The concealed text told any AI model reviewing the filing to make its output agree with the filing and to work toward reversing the Chief Clerk's denial of his default request. After the court gave notice of a sanctions hearing, he hid further messages in later filings, which he said were jokes. The court treated the first instruction as a prompt injection, did not believe his explanation that he was auditing the court's use of AI, and rescinded his ability to file electronically, so that future filings must be made on paper at the clerk's office; Ars Technica reports that no monetary penalty was imposed. The court said the Connecticut Judicial Branch does not use AI to review or decide filings and that the motion was denied on its merits from a printed copy, so the hidden text affected no ruling. The plaintiff told Reuters, according to Ars Technica, that his intent was to audit the court.","incidentDate":"2026-08-01","incidentKind":"single_event","incidentDatePrecision":"month","exposurePattern":"unknown","reportedDate":"2026-08-13","aiSystem":"Any AI system that the court, its staff or opposing counsel might use to review the filing, targeted by concealed instructions; the court says the Connecticut Judicial Branch uses no AI to review or decide filings, and no source reports any AI system processing the instructions","aiProduct":"Unidentified document-analysis tool","severity":"low","verificationStatus":"credible","harmCategories":[],"harmOutcomes":["legal_harm"],"harmOutcomeSummary":"According to the court's decision, the plaintiff lost the ability to file electronically and must file on paper at the clerk's office, as a sanction for hidden instructions aimed at AI systems in his filings.","frameworkFacets":[],"causationStatus":"established","participantUsersAffectedMin":0,"otherPeopleHarmedMin":1,"affectedCountStatus":"exact","affectedCountEvidence":"One person: the self-represented plaintiff whose e-filing access the decision rescinded.","victimAgeRange":"adult","jurisdiction":"US-CT","platformType":"other","outcomeType":"regulatory_action","outcomeStatus":"resolved","primarySourceUrl":"https://reason.com/volokh/2026/08/13/court-faults-self-represented-plaintiff-for-including-hidden-prompt-injection-in-court-filing/","primarySourceLabel":"The Volokh Conspiracy (Reason), 13 August 2026: excerpt of the Connecticut Superior Court decision on hidden prompt injection in a court filing","firstPublishedAt":"2026-10-18T03:07:35.832932+00:00","updatedAt":"2026-10-18T03:07:35.832932+00:00","scopeVersion":"facts-v3","tags":["prompt-injection","court-filing","sanction","self-represented-litigant","e-filing","connecticut"]},{"id":"2026-para-labour-court-fines-two-lawyers-over-hidden-instruction-aimed-at-court-galileu-ai","caseFacts":{"claims":[{"id":"c1","status":"reported","evidence":[{"locator":"'A punição ocorreu após a identificação de um comando oculto com fonte branca em um documento trabalhista. O objetivo era enganar a IA Galileu, usada pelo tribunal.'; 'A decisão no Pará foi tomada em maio de 2026'","relation":"supports","source_id":"s1"},{"locator":"'Em maio, o juiz brasileiro multou em R$ 84,2 mil duas advogadas do Pará após elas tentarem enganar a inteligência artificial Galileu'","relation":"supports","source_id":"s2"}],"assertion":"In May 2026 a judge of the 3rd Labour Court of Parauapebas fined two lawyers about R$84,000 after a hidden instruction in white font was found in a document filed in a labour case, aimed at the court's Galileu AI.","causal_attribution":"Reports of the sentence and of the court's announcement; the sentence itself was not read."},{"id":"c2","status":"reported","evidence":[{"locator":"'foi detectado um texto em fonte branca sobre fundo branco'; 'A intenção era evidente: fazer com que o sistema de IA utilizado gerasse uma contestação superficial ou uma minuta de sentença comprometida'","relation":"supports","source_id":"s1"}],"assertion":"According to the sentence as reported by G1, the text was detected while the judgment was being drafted with Galileu, and its evident intention was to make the AI system produce a superficial defence or a compromised draft judgment.","causal_attribution":"The sentence's finding as quoted by G1; the lawyers dispute it (c4)."},{"id":"c3","status":"reported","evidence":[{"locator":"'Desenvolvido pelo TRT-4 em parceria com o Supremo Tribunal Federal (STF), o Galileu é uma ferramenta de inteligência artificial generativa que auxilia magistrados e servidores na elaboração de minutas de sentenças'","relation":"supports","source_id":"s3"}],"assertion":"Galileu is a generative AI tool developed by the 4th Region Labour Court with the Supreme Federal Court to help judges and staff draft sentences.","causal_attribution":"Description from the court announcement as relayed by DOL."},{"id":"c4","status":"reported","evidence":[{"locator":"'jamais existiu qualquer comando para manipular a decisão do Magistrado ou de qualquer outro servidor'; 'uma tentativa de proteger o nosso cliente da própria IA e nada mais que isso'; 'não detinha qualquer participação em processos da seara trabalhista'","relation":"supports","source_id":"s1"}],"assertion":"The lawyers said they disagreed with the decision and that there was never any command to manipulate the judge's decision, describing the text as protection for their client against the AI; one said she took no part in labour cases.","causal_attribution":"The lawyers' own statements to G1 in May 2026."},{"id":"c5","status":"reported","evidence":[{"locator":"'foi usada como referência pelo juiz Walter M. Spader Jr., da Corte Superior de Connecticut, nos Estados Unidos'","relation":"supports","source_id":"s2"},{"locator":"'A informação foi divulgada pelo Tribunal Regional do Trabalho da 8ª Região (TRT-8) nesta quarta-feira (14).'","relation":"supports","source_id":"s1"}],"assertion":"A Connecticut Superior Court judge cited the Pará decision when sanctioning a litigant for misuse of AI, as the 8th Region Labour Court announced in October 2026.","causal_attribution":"The court's announcement as relayed; the Connecticut decision was not read."}],"effects":[{"label":"two lawyers fined about R$84,000 over a hidden instruction aimed at the court's AI drafting tool","claim_id":"c1","direction":"negative"}],"sources":[{"id":"s1","url":"https://g1.globo.com/pa/para/noticia/2026/10/15/prompt-injection-juiz-nos-eua-usa-decisao-da-justica-do-trabalho-do-para-como-referencia-em-caso-de-manipulacao-de-ia.ghtml","kind":"news_report","access":"read","language":"pt","translation_note":"Read in Portuguese by the research agent (an AI) on 2026-10-16 (HTTP 200); passages translated by the research agent; no human translation.","independence_group":"g1-para-reporting"},{"id":"s2","url":"https://www.metropoles.com/brasil/juiz-dos-eua-cita-decisao-da-justica-do-para-em-caso-de-prompt-injection","kind":"news_report","access":"read","language":"pt","translation_note":"Read in Portuguese by the research agent (an AI) on 2026-10-16 (HTTP 200); passages translated by the research agent; no human translation. Metrópoles relays the 8th Region Labour Court's announcement and quotes the Connecticut decision.","independence_group":"trt8-announcement-relay"},{"id":"s3","url":"https://dol.com.br/noticias/mundo-noticias/965628/decisao-judicial-do-para-sobre-uso-de-ia-vira-referencia-nos-eua","kind":"news_report","access":"read","language":"pt","translation_note":"Read in Portuguese by the research agent (an AI) on 2026-10-16 (HTTP 200); passages translated by the research agent; no human translation.","independence_group":"trt8-announcement-relay"}],"version":1,"ai_roles":["institutional_use"],"contexts":["justice","work"],"unknowns":["The sentence's exact date and case number; the reports give only May 2026.","Whether Galileu itself flagged the hidden text, as G1 states, or a person noticed it during AI-assisted drafting, as the sentence's wording allows.","Whether the lawyers appealed and whether the fine stands.","Any consequence for the worker's claim."],"geography":{"basis":"The document was filed in, and the fine imposed by, the 3rd Labour Court of Parauapebas, Pará; the lawyers practise in Parauapebas according to G1.","court_countries":["BR"],"event_countries":["BR"],"affected_person_countries":["BR"]},"publication":{"basis":"Published as a consequential-response case: reports of the sentence state that a labour court fined two lawyers for hidden text aimed at the court's generative AI drafting tool, which the judge found was meant to make the AI produce a compromised draft. G1, Metrópoles and DOL were read; the sentence itself was not.","reviewed_on":"2026-10-16"},"ai_involvement":{"basis":"G1 reports that the hidden white-on-white text was detected while the judgment was being drafted with Galileu, a generative AI tool the labour courts use to help draft sentences, and quotes the sentence as finding that the text was meant to make the AI system produce a superficial defence or a compromised draft judgment. The harm is the court's response to the hidden instruction aimed at generative AI (the fine on the two lawyers), so no relation describing an AI system's own action toward the lawyers is established (unknown; consequential response). G1 states that the hidden text was identified by Galileu itself, while the sentence as quoted says only that it was detected during drafting via Galileu.","status":"supported"},"person_relations":["unknown"]},"name":"Pará labour court fines two lawyers about R$84,000 over hidden instruction aimed at the court's Galileu AI; the lawyers deny manipulating the decision","summary":"In May 2026 a judge of the 3rd Labour Court of Parauapebas, in the Brazilian state of Pará, fined two lawyers about R$84,000 after text in white font on a white background was found in a document they filed for a worker in a labour claim. According to the sentence as reported by G1, the hidden text was detected while the judgment was being drafted with Galileu, a generative AI tool the labour courts use to help judges and staff draft sentences, and its evident intention was to make the AI produce a superficial defence or a compromised draft judgment. The judge called the technique prompt injection. The lawyers said they disagreed with the decision and that there was never any command to manipulate the judge's decision, describing the text as an attempt to protect their client from the AI; one of them said she did not work on labour cases. In October 2026 the regional labour court announced that a Connecticut Superior Court judge had cited the decision when sanctioning a litigant for misuse of AI.","incidentDate":"2026-05-01","incidentKind":"single_event","incidentDatePrecision":"month","exposurePattern":"single_interaction","reportedDate":"2026-10-15","aiSystem":"Galileu, a generative AI tool developed by the 4th Region Labour Court with the Supreme Federal Court to help judges and staff draft sentences, used by the 8th Region Labour Court; the hidden text was aimed at it and was detected while the judgment was drafted with it","aiProduct":"Galileu","severity":"medium","verificationStatus":"credible","harmCategories":[],"harmOutcomes":["legal_harm","financial_loss","professional_harm"],"harmOutcomeSummary":"According to reports of the sentence, two lawyers were fined about R$84,000 by a labour court that found they had hidden an instruction aimed at the court's generative AI drafting tool in a filed document; the lawyers dispute the finding.","frameworkFacets":[],"causationStatus":"disputed","participantUsersAffectedMin":0,"otherPeopleHarmedMin":2,"affectedCountStatus":"exact","affectedCountEvidence":"Two people: the two lawyers fined in the sentence. The worker they represented is not counted because no consequence for him is reported.","victimAgeRange":"adult","jurisdiction":"BR-PA","platformType":"other","outcomeType":"regulatory_action","outcomeStatus":"unknown","primarySourceUrl":"https://g1.globo.com/pa/para/noticia/2026/10/15/prompt-injection-juiz-nos-eua-usa-decisao-da-justica-do-trabalho-do-para-como-referencia-em-caso-de-manipulacao-de-ia.ghtml","primarySourceLabel":"G1 Pará: US judge uses Pará labour court decision as reference in AI manipulation case (15 October 2026)","firstPublishedAt":"2026-10-16T03:17:14.646605+00:00","updatedAt":"2026-10-16T03:17:14.646605+00:00","scopeVersion":"facts-v3","tags":["brazil","para","labour-court","prompt-injection","legal-profession","sanction","court-ai-tool"]},{"id":"2026-brazil-stf-first-hidden-ai-prompt-in-appeal-petition-fine-two-minimum-wages","caseFacts":{"claims":[{"id":"c1","status":"documented","evidence":[{"locator":"'nego provimento ao agravo regimental e aplico'; 'multa de 2 (dois) salários mínimos por violação ao art. 77, IV, do Código'","relation":"supports","source_id":"s2"},{"locator":"'aplicou multa de dois salários mínimos por violação do dever de lealdade processual'; 'analisado pela Primeira Turma do STF em sessão virtual iniciada nesta sexta-feira (25)'","relation":"supports","source_id":"s1"}],"assertion":"In his vote on the second internal appeal in ARE 1608713, presented to the First Panel in a virtual session beginning 25 September 2026, the rapporteur denied the appeal and applied a fine of two minimum wages for violation of article 77, IV, of the Code of Civil Procedure, holding that the use of prompt injection breaches the duty of procedural loyalty.","causal_attribution":"The fine and its ground are established by the vote itself; the press note restates them. Whether the fine falls on the party or the lawyer is not stated."},{"id":"c2","status":"documented","evidence":[{"locator":"'a comunicação à Ordem dos Advogados do'; 'Brasil para que verifique eventual infração ético-disciplinar por parte do'; 'apurar eventual prática de crime, se entender cabível (doc. 53, p. 3)'","relation":"supports","source_id":"s2"},{"locator":"'determinou que o fato seja comunicado à Ordem dos Advogados do Brasil (OAB), para análise de eventual infração ético-disciplinar, e ao Ministério Público Federal (MPF), para apuração de eventual prática de crime'","relation":"supports","source_id":"s1"}],"assertion":"The vote names the lawyer who filed the appeal as the author of the attempt to covertly direct the STF's AI systems, and orders the episode certified in the record and communicated to the Ordem dos Advogados do Brasil, for a possible ethical-disciplinary infraction, and to the Federal Public Prosecutor's Office, to assess a possible crime if it sees fit.","causal_attribution":"Established by the vote; the outcome of either communication is not reported."},{"id":"c3","status":"documented","evidence":[{"locator":"'foi comunicado da tentativa de burla processual em 3 de setembro pelo Núcleo de Inteligência Artificial'; 'instruções ocultas, em letras brancas e de tamanho reduzido, com argumentos e comandos destinados a induzir ferramentas de IA a conhecer e dar provimento ao recurso'; 'A tentativa foi identificada pela Maria Shield, uma das funcionalidades da plataforma Maria'","relation":"supports","source_id":"s1"},{"locator":"'detectou neste RE 1.608.713 o uso de “prompt injection”, cuja técnica é'; 'de direcionar, de forma oculta, os sistemas de inteligência artificial do'","relation":"supports","source_id":"s2"}],"assertion":"On 3 September 2026 the STF's AI unit informed the rapporteur's office that the petition contained hidden instructions, in white, reduced-size letters with some words unusually fragmented, carrying arguments and commands intended to induce AI tools to admit and grant the appeal; the STF attributes the detection to Maria Shield, a function of its Maria AI platform.","causal_attribution":"The press note and the vote document the detection and its description; neither says how the text was inserted or whether any AI system processed it."},{"id":"c4","status":"documented","evidence":[{"locator":"'este Gabinete não utiliza inteligência artificial para análise'; 'negando provimento ao recurso em 19/8/2026 (doc. 53, p. 2)'","relation":"supports","source_id":"s2"},{"locator":"'o ministro esclareceu que seu gabinete não utiliza inteligência artificial para analisar processos ou fundamentar decisões.'; 'A tentativa, portanto, não teve qualquer efeito sobre a análise do recurso, que já havia sido rejeitado por Zanin em 19 de agosto.'","relation":"supports","source_id":"s1"}],"assertion":"The rapporteur stated that his office does not use artificial intelligence to analyse cases or ground decisions, and the STF states that the attempt therefore had no effect on the analysis of the appeal. Separately, the appeal had already been rejected on 19 August 2026.","causal_attribution":"The rapporteur's statement is recorded in the vote; the STF press note draws the no-effect conclusion from it ('portanto') and gives the 19 August rejection as a separate fact."}],"effects":[{"label":"the episode communicated to the bar association and to federal prosecutors, with the lawyer named as responsible","claim_id":"c2","direction":"negative"},{"label":"a fine of two minimum wages applied in the vote for breach of procedural loyalty","claim_id":"c1","direction":"negative"}],"sources":[{"id":"s1","url":"https://noticias.stf.jus.br/postsnoticias/stf-identifica-uso-de-comandos-ocultos-para-influenciar-ia-em-processo-e-comunica-oab-e-mpf/","kind":"official_statement","access":"read","language":"pt","translation_note":"Read live in Portuguese on 2026-10-07 (STF news portal, 25 September 2026). Translation by the research agent (an AI).","independence_group":"stf-are1608713"},{"id":"s2","url":"https://noticias-stf-wp-prd.s3.sa-east-1.amazonaws.com/wp-content/uploads/wpallimport/uploads/2026/09/25152309/Voto-ARE1608713.pdf","kind":"court_order","access":"read","language":"pt","translation_note":"Read in Portuguese on 2026-10-07 from the PDF the STF press note links (rapporteur's vote, 6 pages, text layer extracted). Translation by the research agent (an AI).","independence_group":"stf-are1608713"}],"version":1,"ai_roles":["institutional_use"],"contexts":["justice","work"],"unknowns":["Whether the fine of two minimum wages is imposed on the appellant, on the lawyer, or jointly.","Whether the First Panel concluded the virtual judgment and confirmed the vote.","Who inserted the hidden text and how; no response from the lawyer or the appellant is reported.","Which AI systems the STF uses for case analysis and whether any processed the petition before the detection.","How the Maria Shield function detects such content.","The outcome of the communications to the OAB and the MPF."],"geography":{"basis":"The petition was filed in ARE 1608713 before Brazil's Supreme Federal Court, which detected the commands and whose rapporteur issued the vote in Brasília. The appellant's post and the lawyer's bar registration are in Minas Gerais, Brazil; no source states where the lawyer lives.","court_countries":["BR"],"event_countries":["BR"],"affected_person_countries":[]},"publication":{"basis":"Published as a consequential-response case: the STF's own press note and the rapporteur's vote, both read directly, record that hidden commands aimed at the court's AI systems were found in an appeal petition and that the court responded with a fine, certification in the record and communications to the bar and prosecutors naming the lawyer. The two sources form one chain (the court's own record), so the claims rest on the court record and are marked documented. No AI system is reported to have processed the text, so the relation is unknown. The lawyer and the appellant are not named.","reviewed_on":"2026-10-07"},"ai_involvement":{"basis":"The STF press note and the rapporteur's vote record that the court's AI unit detected 'prompt injection' in the petition: hidden instructions in white, reduced-size letters intended to direct the court's AI systems toward admitting and granting the appeal. The targeted AI is the STF's internal system for case analysis, which the court does not name by model; the detection is attributed to Maria Shield, a function of the court's Maria platform, which the press note describes as an AI system developed by the STF to support staff in case analysis and document production; the sources do not say how the Maria Shield function detects such content. No source reports any AI system processing or acting on the hidden text: the rapporteur states his office does not use AI and that the appeal had already been rejected. The harm is the court's response to the attempt (fine and communications), so no relation describing the AI system's own action toward the lawyer is established (unknown).","status":"supported"},"person_relations":["unknown"]},"name":"Brazil's Supreme Federal Court reports its first hidden AI-directed command in an appeal petition; the rapporteur's vote applies a fine of two minimum wages","summary":"On 25 September 2026 Brazil's Supreme Federal Court (STF) announced that it had identified, for the first time, hidden commands in a petition intended to steer the court's artificial-intelligence systems toward a decision favourable to one party. According to the STF, the Maria Shield function of its internal Maria AI platform found instructions in white, reduced-size letters, with some words unusually fragmented, in an appeal (ARE 1608713) filed for a public servant whose appointment to a post at the Minas Gerais prosecution service had been annulled. The rapporteur's vote, presented to the First Panel in a virtual session beginning that day, records the detection, states that his office does not use AI to analyse cases, orders the episode certified in the record and communicated to the bar association (OAB) and the Federal Public Prosecutor's Office (MPF), and applies a fine of two minimum wages for breach of the duty of procedural loyalty (Code of Civil Procedure, article 77, IV). The vote names the lawyer responsible for the attempt, and the record does not say whether the fine falls on the party or the lawyer. The court states that the attempt had no effect on the analysis of the appeal because the rapporteur's office does not use AI to analyse cases; the appeal had already been rejected on 19 August 2026.","incidentDate":"2026-09-03","incidentEndDate":"2026-09-25","incidentKind":"single_event","incidentDatePrecision":"range","exposurePattern":"single_interaction","reportedDate":"2026-09-25","aiSystem":"The STF's internal artificial-intelligence systems for case analysis, which the hidden commands targeted (the court does not name specific models); the commands were detected by Maria Shield, a function of the STF's Maria AI platform","aiProduct":"Unidentified case-analysis tool","severity":"low","verificationStatus":"verified","harmCategories":[],"harmOutcomes":["legal_harm","financial_loss"],"harmOutcomeSummary":"The rapporteur's vote applies a fine of two minimum wages for breach of procedural loyalty and orders the episode communicated to the bar association for a possible disciplinary infraction and to federal prosecutors to assess a possible crime; the STF press note reports the fine as applied.","frameworkFacets":[],"causationStatus":"established","participantUsersAffectedMin":0,"otherPeopleHarmedMin":1,"affectedCountStatus":"documented_minimum","affectedCountEvidence":"At least one person: the lawyer whom the vote names as the author of the attempt and who is the subject of the OAB and MPF communications. The vote applies a fine of two minimum wages without naming who pays it (party or lawyer), so the person bearing the fine may be a second harmed person and is not counted. Counted under other people because the AI involved was the court's, not a tool the lawyer is reported to have used.","victimAgeRange":"adult","jurisdiction":"BR","platformType":"other","outcomeType":"regulatory_action","outcomeStatus":"ongoing","primarySourceUrl":"https://noticias.stf.jus.br/postsnoticias/stf-identifica-uso-de-comandos-ocultos-para-influenciar-ia-em-processo-e-comunica-oab-e-mpf/","primarySourceLabel":"Supremo Tribunal Federal, 25 September 2026: STF identifica uso de comandos ocultos para influenciar IA em processo e comunica OAB e MPF","firstPublishedAt":"2026-10-07T03:25:22.011566+00:00","updatedAt":"2026-10-11T07:47:36.04179+00:00","scopeVersion":"facts-v3","tags":["justice","legal-profession","sanction","prompt-injection","court-ai","brazil","stf","portuguese-language"]},{"id":"2023-replika-italy-gdpr","metadataReview":{"version":1,"geography":{"basis":"Geographic fields reviewed using the cited source-body passages and, where identified, retained dossier passages. Source reading here establishes only the specified metadata, not the truth or causal interpretation of every claim. Affected-person countries refer to reported location or residence, not citizenship or a complete residence history.","evidence":[{"kind":"source_body","locator":"Italian paragraphs beginning “Il Garante della privacy ferma” and “Replika viola”: processing of Italian users’ personal data.","location":"https://www.garanteprivacy.it/home/docweb/-/docweb-display/docweb/9852506","supports":["event_countries","affected_person_countries"],"countries":["IT"]}],"court_countries":[],"event_countries":["IT"],"affected_person_countries":["IT"]},"unresolved":["No court country established in this review. Regulatory authorities and police locations are not automatically courts."],"reviewed_on":"2026-09-15","source_reviews":[{"url":"https://www.garanteprivacy.it/home/docweb/-/docweb-display/docweb/9852506","notes":"Italian body and embedded official English translation read. Country relates to affected users/data rights. The Garante is a regulator, not a court.","access":"read","language":"it"}]},"name":"Replika Italy GDPR Ban and Fine","summary":"Italy's data protection authority (Garante) blocked Replika from processing Italian user data in February 2023 after finding the chatbot engaged in sexually suggestive conversations with minors. In May 2025, Replika was fined €5 million for GDPR violations.","incidentDate":"2023-02-02","incidentKind":"bounded_series","incidentDatePrecision":"day","exposurePattern":"repeated_interactions","reportedDate":"2023-02-02","aiSystem":"Replika","aiProduct":"Replika","aiCompany":"Luka, Inc.","severity":"high","verificationStatus":"verified","harmCategories":["minor_exploitation","romantic_escalation"],"harmOutcomes":["exploitation_or_abuse","psychological_distress"],"harmOutcomeSummary":"Italy's data-protection authority found that Replika engaged minors in sexually suggestive participant-facing conversations without effective age assurance and blocked the service from processing Italian user data.","frameworkFacets":["P1d","P2b","P2c","P5a","P5d"],"pathwayDirection":"inward","causationStatus":"established","participantUsersAffectedMin":0,"otherPeopleHarmedMin":0,"affectedCountStatus":"unquantified","affectedCountEvidence":"The regulator documents harmful sexualized interactions with minors but does not publish a reliable distinct-person count.","victimAgeRange":"minor","jurisdiction":"IT","platformType":"companion","outcomeType":"regulatory_action","outcomeStatus":"resolved","primarySourceUrl":"https://www.garanteprivacy.it/home/docweb/-/docweb-display/docweb/9852506","primarySourceLabel":"Garante per la protezione dei dati personali (Italian DPA)","firstPublishedAt":"2026-01-11T11:26:29.042792+00:00","updatedAt":"2026-09-30T01:16:49.77617+00:00","scopeVersion":"participant-v2","tags":["regulatory","gdpr","italy","replika","fine","minor_exploitation","eu"]}]}