{"meta":{"exportedAt":"2026-10-31T07:56:14.517Z","formatVersion":2,"selection":{"q":"blog","system":"","harm":"","context":"","country":"","role":"","relation":"","evidence":"","year":"unknown","response":"","severity":"","verification":"","view":"incidents","sort":"added"},"totalIncidents":3,"coverage":{"cases":3,"countries":0,"languages":1,"unknownLocation":3,"locationPending":0,"unknownLanguage":0,"unknownDate":3,"lawsuits":0,"regulatory":0,"minors":0,"coreRelations":2,"contextualRelations":1,"mixedRelations":0,"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":"2026-patient-says-doctor-read-google-ai-search-summary-for-a-diagnosis-then-called-to-say-the-patient-was-right-first-person","caseFacts":{"claims":[{"id":"c1","status":"reported","evidence":[{"locator":"'one of my appointments, where a doctor pulled up Google on his computer, said, “Let’s ask Dr. Google,” and read the summary at the top of the page without clicking through to any of the sources'","relation":"supports","source_id":"s1"}],"assertion":"The author wrote that at one of their appointments a doctor pulled up Google on his computer, said \"Let's ask Dr. Google,\" and read the summary at the top of the page without clicking through to any of the sources.","causal_attribution":"Author's first-person account of the appointment."},{"id":"c2","status":"reported","evidence":[{"locator":"'Google a diagnosis, and read me the AI summary at the top of the results'; 'Spoiler alert: He was actually wrong.'","relation":"supports","source_id":"s2"}],"assertion":"In a Bluesky post linking the article, the author described the doctor as Googling a diagnosis and reading them the AI summary at the top of the results, and wrote that he was actually wrong.","causal_attribution":"The author identifies the summary as AI-generated and the doctor's answer as wrong; what the summary said is not described, so whether the error lay in the summary itself or in the doctor's reading of it is not established."},{"id":"c3","status":"reported","evidence":[{"locator":"'To his credit, he did call me to say I was right after I emailed to push back and essentially say, “I think you’re wrong, here are my sources. What am I missing?”'","relation":"supports","source_id":"s1"}],"assertion":"The author wrote that the doctor called to say they were right after they emailed to push back, saying in effect that he was wrong and giving their sources.","causal_attribution":"Author's account of the correction."},{"id":"c4","status":"reported","evidence":[{"locator":"'I’d be 100% lying if I said that I wasn’t nervous to hit send on the emails to the doctors to question their diagnoses'","relation":"supports","source_id":"s1"}],"assertion":"The author wrote that they were nervous to send the emails questioning the doctors' diagnoses.","causal_attribution":"The author describes this for both diagnosis disputes in the article, only one of which involved the AI summary."},{"id":"c5","status":"reported","evidence":[{"locator":"'I definitely used AI when I was researching this diagnosis. But I didn’t just go with the first response Claude gave me'; 'I used Gemini to do deeper research and had it try to find holes in what Claude told me'","relation":"supports","source_id":"s1"}],"assertion":"The author wrote that they used Claude and Gemini to research the diagnosis, cross-checked them against other sources and read the cited sources themselves.","causal_attribution":"Context: the author's own AI-assisted research is how they challenged the doctor."}],"effects":[{"label":"a doctor's answer on a diagnosis, read from an AI search summary, that the doctor later agreed was wrong (author's account)","claim_id":"c3","direction":"negative"},{"label":"nervousness at emailing the doctors to question their diagnoses (author's account)","claim_id":"c4","direction":"negative"},{"label":"the author researched the diagnosis with Claude and Gemini and read the cited sources (author's account)","claim_id":"c5","direction":"neutral"}],"sources":[{"id":"s1","url":"https://medium.com/design-bootcamp/lets-ask-dr-google-blind-trust-ai-baba28829c6a","kind":"personal_blog","access":"read","language":"en","translation_note":"Read in English on 2026-10-31. medium.com returned HTTP 403 to the fetcher and the Wayback Machine had no capture, so the full article text was read from the author's own Medium RSS feed (HTTP 200), which carries the complete article dated 26 October 2026.","independence_group":"author-dr-google-account"},{"id":"s2","url":"https://bsky.app/profile/sarahdoody.bsky.social/post/3mz3w5eavmk23","kind":"social_post","access":"read","language":"en","translation_note":"Read in English on 2026-10-31 through the public Bluesky API (post, thread with no replies, and the author's feed, HTTP 200). Same author as the article, so not independent of it.","independence_group":"author-dr-google-account"}],"version":1,"ai_roles":["institutional_use","own_use"],"contexts":["health"],"unknowns":["What the AI summary said, what diagnosis was at issue, and whether the error lay in the summary or in the doctor's reading of it.","When the appointment took place; the article (26 October 2026) gives no date, and the Bluesky post of 30 October 2026 does not either.","Which Google feature produced the summary; the author calls it the AI summary at the top of the results.","Whether the wrong answer affected any treatment before it was corrected.","Where the appointment took place."],"geography":{"basis":"The article and the post do not say where the appointment took place or where the author lives.","court_countries":[],"event_countries":[],"affected_person_countries":[]},"publication":{"basis":"Published as a contextual case: the author says a doctor answered a diagnosis question by reading Google's AI search summary aloud during the appointment, that the answer was wrong, and that the doctor agreed after the author pushed back with sources. First-person account, low severity.","reviewed_on":"2026-10-31"},"ai_involvement":{"basis":"The author writes that during their appointment a doctor searched Google for a diagnosis and read them the AI-generated summary at the top of the results without opening any sources, and that the doctor's answer was wrong; after the author emailed sources, the doctor called to say the author was right. The output was used in the doctor's answer to the patient. What the summary said is not described.","status":"reported"},"person_relations":["informed_decision_about"]},"name":"First-person account: patient says a doctor read Google's AI search summary for a diagnosis and later called to say the patient was right","summary":"In an article published on Medium on 26 October 2026 and a Bluesky post on 30 October, a writer says that at one appointment a doctor said \"Let's ask Dr. Google\", searched for a diagnosis and read the AI summary at the top of the results without opening any sources. The writer says the doctor was wrong; after they emailed him their sources and asked what they were missing, he called to say they were right. They say they were nervous to send the emails, and that their own research used Claude and Gemini, cross-checked against the cited sources. The article does not say what the summary said, what the diagnosis was, or when the appointment took place.","incidentKind":"single_event","incidentDatePrecision":"unknown","exposurePattern":"single_interaction","reportedDate":"2026-10-26","aiSystem":"AI-generated summary at the top of Google search results, as described by the patient; feature name not given","aiProduct":"Google Search AI summary (reported)","severity":"low","verificationStatus":"unverified","harmCategories":[],"harmOutcomes":["psychological_distress"],"harmOutcomeSummary":"The author says a doctor gave them a wrong answer on a diagnosis after reading Google's AI search summary aloud, and that they had to research and nervously challenge it by email before the doctor agreed (first-person account; no effect on treatment is reported).","frameworkFacets":[],"causationStatus":"alleged","participantUsersAffectedMin":0,"otherPeopleHarmedMin":1,"affectedCountStatus":"exact","affectedCountEvidence":"One person counted: the author, the patient about whom the doctor used the AI summary; the author was not the AI user.","victimAgeRange":"unknown","platformType":"assistant","outcomeStatus":"resolved","primarySourceUrl":"https://medium.com/design-bootcamp/lets-ask-dr-google-blind-trust-ai-baba28829c6a","primarySourceLabel":"Medium (Bootcamp), 26 Oct 2026: \"Let’s ask Dr. Google: The risk of blind trust in AI\"","firstPublishedAt":"2026-10-31T03:25:14.801466+00:00","updatedAt":"2026-10-31T03:25:14.801466+00:00","scopeVersion":"facts-v3","tags":["first-person","blog","bluesky","health","clinician","ai-search-summary"]},{"id":"2026-user-says-claude-read-connected-email-in-auto-approve-session-despite-always-ask-setting-first-person","caseFacts":{"claims":[{"id":"c1","status":"reported","evidence":[{"locator":"'I found out a few days ago, that regardless of your permission settings on connectors, Claude auto-mode is able to use them.'; 'what can you tell me about [platform.com]'; 'the service provider that emailed you at 1:59pm today likely uses [platform] to manage projects'","relation":"supports","source_id":"s1"}],"assertion":"The author says that a few days before the post of 20 October 2026, Claude answered a question about a platform named in an email by saying that the service provider that had emailed the author at 1:59pm that day likely used the platform.","causal_attribution":"The author's account; Claude's quoted answer is the evidence that the assistant drew on the email."},{"id":"c2","status":"reported","evidence":[{"locator":"'all actions under the Fastmail connector were set to \"Always ask for permission\".'; 'They responded promptly that this was expected behavior'; 'Claude asks before each use, except in sessions set to Automatically approve'","relation":"supports","source_id":"s1"}],"assertion":"The author says the Fastmail connector's actions were all set to always ask for permission, and that Anthropic replied that use in sessions set to automatically approve was expected behaviour, pointing to a tooltip saying Claude asks before each use except in such sessions.","causal_attribution":"The author's account of the setting and of the vendor's reply; the reply itself was not inspected."},{"id":"c3","status":"reported","evidence":[{"locator":"'This violated my expectations, and exposed me to'; 'email is very sensitive (even read only)'; 'removed the fastmail connector from Claude'","relation":"supports","source_id":"s1"}],"assertion":"The author says the behaviour violated their expectations and exposed them to risk, that they are not ready to trust auto mode with email, and that they removed the Fastmail connector from Claude.","causal_attribution":"The author connects the experience and the removal of the connector to Claude's unrequested use of the email connector."}],"effects":[{"label":"the assistant read the user's connected email without asking, contrary to the user's always-ask setting, and the user withdrew the connection","claim_id":"c3","direction":"negative"}],"sources":[{"id":"s1","url":"https://blog.baldassari.me/auto-mode-can-use-connectors/","kind":"first_person_account","access":"read","language":"en","translation_note":"Read in English on 2026-10-21 from the author's blog; no translation was needed and the research agent (an AI) read the text directly. The post is dated 20 October 2026 and carries no screenshots.","independence_group":"author-blog-claude-connector"}],"version":1,"ai_roles":["own_use"],"contexts":["privacy","everyday_life"],"unknowns":["The date of the exchange beyond 'a few days ago' before 20 October 2026.","Which Claude product surface was used and how the session came to be set to automatically approve.","The text of Anthropic's reply, which the author paraphrases."],"geography":{"basis":"The post gives no location for the user or the exchange; no country is inferred from the blog's domain or language.","court_countries":[],"event_countries":[],"affected_person_countries":[]},"publication":{"basis":"A concrete first-person account of an AI assistant using the user's connected email without asking, contrary to the user's permission setting, with the assistant's own answer quoted as evidence and an adverse experience (loss of control over access to private email, withdrawal of the connection) described by the user. All claims are attributed to the author.","reviewed_on":"2026-10-21"},"ai_involvement":{"basis":"The post quotes Claude's own answer, which stated that the service provider that had emailed the user at 1:59pm that day likely used the platform; the user says that detail could only come from the Fastmail connector, whose actions were all set to always ask for permission, and Claude had not asked. The user connects that unrequested use of the email connector to the adverse experience and to removing the connector.","status":"reported"},"person_relations":["acted_on_behalf"]},"name":"User says Claude read their connected email without asking in an automatically approved session although the connector was set to always ask; first-person","summary":"In a personal blog post dated 20 October 2026, a Claude user says that a few days earlier they asked Claude what it could tell them about a platform named in an email from a service provider, and Claude's answer said that the service provider that had emailed them at 1:59pm that day probably used the platform. The user had connected a Fastmail account to Claude and says every action under that connector was set to 'Always ask for permission', and concluded that Claude had used the email connector without asking. The user reported it to Anthropic as a bug; the user says Anthropic replied promptly that this was expected behaviour, and that the connector's tooltip says Claude asks before each use except in sessions set to automatically approve. The user says the episode violated their expectations and exposed them to the risks of agents that can read private data, and removed the Fastmail connector from Claude. The exact date is not given beyond 'a few days ago'.","incidentKind":"single_event","incidentDatePrecision":"unknown","exposurePattern":"single_interaction","reportedDate":"2026-10-20","aiSystem":"Claude (Anthropic) with a Fastmail email connector, used in a session set to automatically approve actions (auto mode), per the user's post","aiProduct":"Claude (reported)","aiCompany":"Anthropic","severity":"low","verificationStatus":"unverified","harmCategories":[],"harmOutcomes":["loss_of_autonomy"],"harmOutcomeSummary":"The author of a personal blog post says Claude read their connected email to answer an unrelated question without asking, although every action of the email connector was set to always ask for permission; the author says this violated their expectations and exposed them to risk, and removed the connector. Anthropic, the author says, replied that this was expected behaviour in sessions set to automatically approve.","frameworkFacets":[],"causationStatus":"alleged","participantUsersAffectedMin":1,"otherPeopleHarmedMin":0,"affectedCountStatus":"exact","affectedCountEvidence":"One person: the author of the post, whose connected email Claude read. Exact 1.","victimAgeRange":"unknown","jurisdiction":"unknown","platformType":"assistant","outcomeType":"internal_action","outcomeStatus":"resolved","primarySourceUrl":"https://blog.baldassari.me/auto-mode-can-use-connectors/","primarySourceLabel":"Personal blog post, 20 October 2026: Claude's \"Always ask\" does not","firstPublishedAt":"2026-10-21T03:09:47.44438+00:00","updatedAt":"2026-10-21T03:09:47.44438+00:00","scopeVersion":"facts-v3","tags":["ai-assistant","email-connector","permission-settings","auto-approve","first-person"]},{"id":"2026-delta-concierge-ai-cancelled-return-flight-after-saying-it-stayed-booked","caseFacts":{"claims":[{"id":"c1","status":"reported","evidence":[{"locator":"canceled my flight home from LAX after saying that it wouldn’t! Now customer service says they can’t help and are expecting me to pay almost twice as much for just a one way to get home","relation":"supports","source_id":"s2"}],"assertion":"The passenger posted that Delta's AI Concierge cancelled her flight home from Los Angeles after saying it would not, and that customer service said they could not help and expected her to pay almost twice as much for a one-way ticket.","causal_attribution":"The passenger's account."},{"id":"c2","status":"reported","evidence":[{"locator":"Your JFK to LAX flight has been canceled, and your LAX to JFK flight remains booked.","relation":"supports","source_id":"s1"},{"locator":"It lists both DL752 from JFK to Los Angeles and DL938 from Los Angeles to JFK.","relation":"supports","source_id":"s1"}],"assertion":"The Concierge told her that the outbound flight had been cancelled and the return flight remained booked, while the refund record she posted listed both flights.","causal_attribution":"The blog's reading of the passenger's screenshots; the screenshots were not inspected."},{"id":"c3","status":"corroborated","evidence":[{"locator":"The team identified a bug that caused the return segment to be unintentionally canceled when a customer attempted a partial-trip cancellation, which Delta Concierge does not currently support (only full-itinerary cancellations).","relation":"supports","source_id":"s1"},{"locator":"canceled my flight home from LAX after saying that it wouldn’t!","relation":"supports","source_id":"s2"}],"assertion":"The return segment was unintentionally cancelled when she attempted a partial-trip cancellation through the Concierge.","causal_attribution":"Delta's statement attributes the cancellation to a bug in the Concierge's handling of a partial cancellation; the passenger independently reports the cancellation."},{"id":"c4","status":"reported","evidence":[{"locator":"Once identified, the customer’s original itinerary was quickly restored, and there was no disruption to her planned travel. We also corrected the functionality and updated Concierge guidance to prevent this from recurring.","relation":"supports","source_id":"s1"}],"assertion":"Delta said her original itinerary was quickly restored with no disruption to her planned travel, and that it corrected the functionality and updated Concierge guidance.","causal_attribution":"Delta's statement."}],"effects":[{"label":"return flight cancelled by the AI assistant while it said the flight stayed booked","claim_id":"c3","direction":"negative"},{"label":"told to pay nearly double for a one-way replacement before the itinerary was restored (passenger account)","claim_id":"c1","direction":"negative"}],"sources":[{"id":"s1","url":"https://viewfromthewing.com/deltas-ai-told-a-passenger-her-flight-home-was-safe-then-canceled-it-and-demanded-double-to-rebook/","kind":"news_report","access":"read","language":"en","translation_note":"Read in English in full from viewfromthewing.com (HTTP 200). Travel blog relaying the passenger's post and screenshots and carrying Delta's statement to the blog; the screenshots were described, not inspected.","independence_group":"viewfromthewing"},{"id":"s2","url":"https://x.com/LAWL_een/status/2084385766487036157","kind":"social_media_post","access":"read","language":"en","translation_note":"Read in English via the public X syndication and oEmbed endpoints (HTTP 200); the passenger's own post. The attached image was not inspected.","independence_group":"passenger-x-post"}],"version":1,"ai_roles":["institutional_use"],"contexts":["everyday_life"],"unknowns":["The date of the exchange, beyond being on or before 3 August 2026.","How long the return segment stayed cancelled before Delta restored it.","The passenger's full request to the Concierge."],"geography":{"basis":"The trip ran between New York (JFK) and Los Angeles, but no source states where the passenger was when she used the in-app assistant or where she lives, so the countries are left unknown.","court_countries":[],"event_countries":[],"affected_person_countries":[]},"publication":{"basis":"Published as a concrete account of an airline's AI assistant cancelling a booked flight it said it would keep, confirmed as a bug by the airline. The adverse effect was reversed when Delta restored the itinerary; the passenger is not named.","reviewed_on":"2026-10-06"},"ai_involvement":{"basis":"Delta's statement says a bug caused the return segment to be unintentionally cancelled when the passenger attempted a partial-trip cancellation, which Delta Concierge does not currently support; the passenger says the Concierge cancelled her flight home, and the blog reports the Concierge told her the return flight remained booked while the refund record listed both flights. The passenger reports that the cancellation left her facing a one-way fare nearly double the cost until Delta restored the itinerary.","status":"supported"},"person_relations":["communicated_with","acted_on_behalf"]},"name":"Delta's Concierge AI cancelled a passenger's return flight that, she says, it had confirmed would stay booked; Delta blamed a bug and restored the trip","summary":"A Delta Platinum Medallion member posted on X on 3 August 2026 that Delta's AI Concierge in the Fly Delta app cancelled her flight home from Los Angeles after saying it would not, and that customer service said they could not help and expected her to pay almost twice as much for a one-way ticket. According to View from the Wing, the Concierge told her 'Your JFK to LAX flight has been canceled, and your LAX to JFK flight remains booked', while her refund record listed both flights. Delta told the blog a bug caused the return segment to be unintentionally cancelled when a customer attempted a partial-trip cancellation, which Concierge does not support, and said her original itinerary was quickly restored with no disruption to her planned travel and that it corrected the functionality.","incidentKind":"single_event","incidentDatePrecision":"unknown","exposurePattern":"single_interaction","reportedDate":"2026-08-03","aiSystem":"Delta Concierge, a generative AI assistant in the Fly Delta app that can cancel trips, in beta","aiProduct":"Delta Concierge","aiCompany":"Delta Air Lines","severity":"low","verificationStatus":"credible","harmCategories":[],"harmOutcomes":["other_material_harm"],"harmOutcomeSummary":"The airline's AI assistant cancelled the passenger's return flight while, she says, telling her it remained booked, and reservations initially quoted nearly double for a one-way replacement before Delta restored the itinerary (passenger's post; Delta's statement confirms the unintended cancellation).","frameworkFacets":[],"causationStatus":"supported","participantUsersAffectedMin":1,"otherPeopleHarmedMin":0,"affectedCountStatus":"exact","affectedCountEvidence":"One person: the passenger whose return segment the Concierge cancelled. No other affected person is described.","victimAgeRange":"adult","platformType":"agent","outcomeType":"internal_action","outcomeStatus":"resolved","primarySourceUrl":"https://viewfromthewing.com/deltas-ai-told-a-passenger-her-flight-home-was-safe-then-canceled-it-and-demanded-double-to-rebook/","primarySourceLabel":"View from the Wing, 4 August 2026: Delta's AI told a passenger her flight home was safe, then canceled it","firstPublishedAt":"2026-10-06T03:22:02.233131+00:00","updatedAt":"2026-10-06T03:22:02.233131+00:00","scopeVersion":"facts-v3","tags":["ai-agent","airline","booking-cancellation","customer-service"]}]}