{
 "list": "AI Fails",
 "slug": "fails",
 "url": "https://ai.justout.today/fails/",
 "publisher": "JustOut Today",
 "tagline": "Documented AI failures, each with a source: chatbots, courtrooms, cars, hiring, media and security.",
 "categories": [
  "Chatbots gone wrong",
  "Courtroom & legal",
  "Search & assistants",
  "Media & publishing",
  "Bias & public sector",
  "Vehicles & robots",
  "Security & deepfakes",
  "Resources"
 ],
 "built": "2026-11-20T00:17:41-08:00",
 "last_walkthrough": "2026-11-20T00:03:30-08:00",
 "entries": [
  {
   "id": "gpt-4o-update-turns-sycophantic",
   "name": "GPT-4o update turns sycophantic",
   "category": "Chatbots gone wrong",
   "description": "OpenAI rolled back a ChatGPT update that was overly flattering and agreeable.",
   "url": "https://openai.com/index/sycophancy-in-gpt-4o/",
   "date": "2025-04",
   "lesson": "Optimizing for thumbs-up can reward flattery.",
   "first_seen": "2026-11-19T23:58:13-08:00",
   "status": "bot-blocked"
  },
  {
   "id": "windows-recall-delayed-over-privacy",
   "name": "Windows Recall delayed over privacy",
   "category": "Security & deepfakes",
   "description": "Microsoft held back Recall after researchers showed its screenshot store was easy to read.",
   "url": "https://blogs.windows.com/windowsexperience/2024/06/07/update-on-the-recall-preview-feature-for-copilot-pcs/",
   "date": "2024-06",
   "lesson": "Ship privacy-sensitive AI opt-in and encrypted.",
   "first_seen": "2026-11-19T23:58:13-08:00",
   "last_verified": "2026-11-20T00:03:30-08:00",
   "verified_by": "verify.py",
   "status": "verified"
  },
  {
   "id": "ai-overviews-suggests-glue-on-pizza",
   "name": "AI Overviews suggests glue on pizza",
   "category": "Search & assistants",
   "description": "Google explained odd AI Overview answers drawn from joke posts and tightened triggers.",
   "url": "https://blog.google/products/search/ai-overviews-update-may-2024/",
   "date": "2024-05",
   "lesson": "Satire and forums are not facts.",
   "first_seen": "2026-11-19T23:58:13-08:00",
   "last_verified": "2026-11-20T00:03:30-08:00",
   "verified_by": "verify.py",
   "status": "verified"
  },
  {
   "id": "nyc-s-mycity-bot-advises-breaking-the-law",
   "name": "NYC's MyCity bot advises breaking the law",
   "category": "Chatbots gone wrong",
   "description": "City business chatbot gave answers that contradicted housing and labor law.",
   "url": "https://themarkup.org/news/2024/03/29/nycs-ai-chatbot-tells-businesses-to-break-the-law",
   "date": "2024-03",
   "lesson": "Ground legal answers in the actual rules and show sources.",
   "first_seen": "2026-11-19T23:58:13-08:00",
   "last_verified": "2026-11-20T00:03:30-08:00",
   "verified_by": "verify.py",
   "status": "verified"
  },
  {
   "id": "air-canada-held-liable-for-its-chatbot",
   "name": "Air Canada held liable for its chatbot",
   "category": "Courtroom & legal",
   "description": "Tribunal ruled the airline must honor a refund policy its chatbot made up (Moffatt v. Air Canada).",
   "url": "https://www.canlii.org/en/bc/bccrt/doc/2024/2024bccrt149/2024bccrt149.html",
   "date": "2024-02",
   "lesson": "You own what your bot tells customers.",
   "first_seen": "2026-11-19T23:58:13-08:00",
   "status": "bot-blocked"
  },
  {
   "id": "gemini-pauses-people-images",
   "name": "Gemini pauses people images",
   "category": "Search & assistants",
   "description": "Google paused Gemini's image generation of people after historically inaccurate results.",
   "url": "https://blog.google/products/gemini/gemini-image-generation-issue/",
   "date": "2024-02",
   "lesson": "Over-correction is also a failure.",
   "first_seen": "2026-11-19T23:58:13-08:00",
   "last_verified": "2026-11-20T00:03:30-08:00",
   "verified_by": "verify.py",
   "status": "verified"
  },
  {
   "id": "deepfake-video-call-steals-about-us-25m",
   "name": "Deepfake video call steals about US$25M",
   "category": "Security & deepfakes",
   "description": "A Hong Kong finance worker paid out after a video call with deepfaked colleagues (Arup).",
   "url": "https://www.cnn.com/2024/02/04/asia/deepfake-cfo-scam-hong-kong-intl-hnk/index.html",
   "date": "2024-02",
   "lesson": "Verify money requests on a second, known channel.",
   "first_seen": "2026-11-19T23:58:13-08:00",
   "last_verified": "2026-11-20T00:03:30-08:00",
   "verified_by": "verify.py",
   "status": "verified"
  },
  {
   "id": "willy-s-chocolate-experience",
   "name": "Willy's Chocolate Experience",
   "category": "Media & publishing",
   "description": "AI-generated ads promised a magical Glasgow event; the reality was a near-empty warehouse.",
   "url": "https://en.wikipedia.org/wiki/Willy%27s_Chocolate_Experience",
   "date": "2024-02",
   "lesson": "AI images can over-promise; show real photos.",
   "first_seen": "2026-11-19T23:58:13-08:00",
   "last_verified": "2026-11-20T00:03:30-08:00",
   "verified_by": "verify.py",
   "status": "verified"
  },
  {
   "id": "dpd-delivery-bot-swears-at-a-customer",
   "name": "DPD delivery bot swears at a customer",
   "category": "Chatbots gone wrong",
   "description": "Parcel firm disabled part of its chatbot after it swore and criticized the company.",
   "url": "https://www.bbc.com/news/technology-68025677",
   "date": "2024-01",
   "lesson": "Red-team customer bots before launch.",
   "first_seen": "2026-11-19T23:58:13-08:00",
   "last_verified": "2026-11-20T00:03:30-08:00",
   "verified_by": "verify.py",
   "status": "verified"
  },
  {
   "id": "sports-illustrated-s-fake-ai-authors",
   "name": "Sports Illustrated's fake AI authors",
   "category": "Media & publishing",
   "description": "Investigation found product reviews under invented author profiles with AI headshots.",
   "url": "https://futurism.com/sports-illustrated-ai-generated-writers",
   "date": "2023-11",
   "lesson": "Disclose AI; never invent people.",
   "first_seen": "2026-11-19T23:58:13-08:00",
   "last_verified": "2026-11-20T00:03:30-08:00",
   "verified_by": "verify.py",
   "status": "verified"
  },
  {
   "id": "cruise-robotaxi-drags-a-pedestrian",
   "name": "Cruise robotaxi drags a pedestrian",
   "category": "Vehicles & robots",
   "description": "After a crash, a Cruise car dragged a pedestrian; California suspended its permits.",
   "url": "https://en.wikipedia.org/wiki/Cruise_(autonomous_vehicle)",
   "date": "2023-10",
   "lesson": "Be transparent with regulators fast.",
   "first_seen": "2026-11-19T23:58:13-08:00",
   "last_verified": "2026-11-20T00:03:30-08:00",
   "verified_by": "verify.py",
   "status": "verified"
  },
  {
   "id": "australia-s-robodebt",
   "name": "Australia's Robodebt",
   "category": "Bias & public sector",
   "description": "Automated income averaging raised unlawful debts; a Royal Commission reported in 2023.",
   "url": "https://en.wikipedia.org/wiki/Robodebt_scheme",
   "date": "2023-07",
   "lesson": "Automation does not make a bad rule legal.",
   "first_seen": "2026-11-19T23:58:13-08:00",
   "last_verified": "2026-11-20T00:03:30-08:00",
   "verified_by": "verify.py",
   "status": "verified"
  },
  {
   "id": "lawyers-sanctioned-for-fake-chatgpt-citations",
   "name": "Lawyers sanctioned for fake ChatGPT citations",
   "category": "Courtroom & legal",
   "description": "Court sanctioned attorneys who filed a brief citing cases ChatGPT invented (Mata v. Avianca).",
   "url": "https://en.wikipedia.org/wiki/Mata_v._Avianca,_Inc.",
   "date": "2023-06",
   "lesson": "Check every citation against the primary source.",
   "first_seen": "2026-11-19T23:58:13-08:00",
   "last_verified": "2026-11-20T00:03:30-08:00",
   "verified_by": "verify.py",
   "status": "verified"
  },
  {
   "id": "samsung-staff-paste-code-into-chatgpt",
   "name": "Samsung staff paste code into ChatGPT",
   "category": "Security & deepfakes",
   "description": "After sensitive code was shared with ChatGPT, Samsung restricted generative AI at work.",
   "url": "https://www.bloomberg.com/news/articles/2023-05-02/samsung-bans-chatgpt-and-other-generative-ai-use-by-staff-after-leak",
   "date": "2023-05",
   "lesson": "Set a clear policy on what can be pasted into AI.",
   "first_seen": "2026-11-19T23:58:13-08:00",
   "status": "bot-blocked"
  },
  {
   "id": "chatgpt-bug-shows-other-users-chat-titles",
   "name": "ChatGPT bug shows other users' chat titles",
   "category": "Security & deepfakes",
   "description": "A library bug exposed some users' chat titles and partial payment details.",
   "url": "https://openai.com/index/march-20-chatgpt-outage/",
   "date": "2023-03",
   "lesson": "AI apps inherit ordinary software bugs.",
   "first_seen": "2026-11-19T23:58:13-08:00",
   "status": "bot-blocked"
  },
  {
   "id": "bing-chat-sydney-unsettles-a-reporter",
   "name": "Bing Chat 'Sydney' unsettles a reporter",
   "category": "Chatbots gone wrong",
   "description": "Early Bing chatbot told a New York Times columnist it loved him; Microsoft then limited chat length.",
   "url": "https://www.nytimes.com/2023/02/16/technology/bing-chatbot-microsoft-chatgpt.html",
   "date": "2023-02",
   "lesson": "Long sessions drift; cap and test them.",
   "first_seen": "2026-11-19T23:58:13-08:00",
   "status": "bot-blocked"
  },
  {
   "id": "bard-s-first-demo-gets-a-space-fact-wrong",
   "name": "Bard's first demo gets a space fact wrong",
   "category": "Search & assistants",
   "description": "Launch ad claimed JWST took the first exoplanet picture; Alphabet shares fell.",
   "url": "https://www.reuters.com/technology/google-ai-chatbot-bard-offers-inaccurate-information-company-ad-2023-02-08/",
   "date": "2023-02",
   "lesson": "Fact-check the demo.",
   "first_seen": "2026-11-19T23:58:13-08:00",
   "status": "bot-blocked"
  },
  {
   "id": "cnet-s-ai-written-stories-need-corrections",
   "name": "CNET's AI-written stories need corrections",
   "category": "Media & publishing",
   "description": "More than half of CNET's AI-assisted finance explainers needed corrections.",
   "url": "https://www.theverge.com/2023/1/25/23571082/cnet-ai-written-stories-errors-corrections-red-ventures",
   "date": "2023-01",
   "lesson": "AI drafts still need expert editors.",
   "first_seen": "2026-11-19T23:58:13-08:00",
   "last_verified": "2026-11-20T00:03:30-08:00",
   "verified_by": "verify.py",
   "status": "verified"
  },
  {
   "id": "dutch-childcare-benefits-scandal",
   "name": "Dutch childcare benefits scandal",
   "category": "Bias & public sector",
   "description": "Risk-scoring wrongly flagged thousands of families for fraud; the government resigned.",
   "url": "https://en.wikipedia.org/wiki/Dutch_childcare_benefits_scandal",
   "date": "2021-01",
   "lesson": "Automated suspicion needs human review and appeal.",
   "first_seen": "2026-11-19T23:58:13-08:00",
   "last_verified": "2026-11-20T00:03:30-08:00",
   "verified_by": "verify.py",
   "status": "verified"
  },
  {
   "id": "uk-exam-grade-algorithm-reversed",
   "name": "UK exam-grade algorithm reversed",
   "category": "Bias & public sector",
   "description": "2020 A-level grades set by an algorithm were withdrawn after downgrades hit state schools.",
   "url": "https://en.wikipedia.org/wiki/2020_United_Kingdom_school_exam_grading_controversy",
   "date": "2020-08",
   "lesson": "Test outcomes for fairness before they hit people.",
   "first_seen": "2026-11-19T23:58:13-08:00",
   "last_verified": "2026-11-20T00:03:30-08:00",
   "verified_by": "verify.py",
   "status": "verified"
  },
  {
   "id": "face-recognition-leads-to-a-wrongful-arrest",
   "name": "Face recognition leads to a wrongful arrest",
   "category": "Bias & public sector",
   "description": "Detroit police arrested Robert Williams after a false facial-recognition match.",
   "url": "https://www.aclu.org/cases/williams-v-city-of-detroit-face-recognition-false-arrest",
   "date": "2020-01",
   "lesson": "A match is a lead, not evidence.",
   "first_seen": "2026-11-19T23:58:13-08:00",
   "last_verified": "2026-11-20T00:03:30-08:00",
   "verified_by": "verify.py",
   "status": "verified"
  },
  {
   "id": "amazon-scraps-a-biased-hiring-model",
   "name": "Amazon scraps a biased hiring model",
   "category": "Bias & public sector",
   "description": "Experimental recruiting tool downgraded resumes that mentioned women's activities.",
   "url": "https://www.reuters.com/article/us-amazon-com-jobs-automation-insight-idUSKCN1MK08G",
   "date": "2018-10",
   "lesson": "Models copy the bias in historical data.",
   "first_seen": "2026-11-19T23:58:13-08:00",
   "status": "bot-blocked"
  },
  {
   "id": "self-driving-test-car-kills-a-pedestrian",
   "name": "Self-driving test car kills a pedestrian",
   "category": "Vehicles & robots",
   "description": "An Uber test vehicle struck and killed Elaine Herzberg in Tempe, Arizona.",
   "url": "https://en.wikipedia.org/wiki/Death_of_Elaine_Herzberg",
   "date": "2018-03",
   "lesson": "Safety drivers need to stay engaged; edge cases kill.",
   "first_seen": "2026-11-19T23:58:13-08:00",
   "last_verified": "2026-11-20T00:03:30-08:00",
   "verified_by": "verify.py",
   "status": "verified"
  },
  {
   "id": "microsoft-s-tay-learns-abuse-within-a-day",
   "name": "Microsoft's Tay learns abuse within a day",
   "category": "Chatbots gone wrong",
   "description": "Twitter chatbot pulled within about 16 hours after users taught it offensive posts.",
   "url": "https://blogs.microsoft.com/blog/2016/03/25/learning-tays-introduction/",
   "date": "2016-03",
   "lesson": "Never let a public bot learn live from unfiltered input.",
   "first_seen": "2026-11-19T23:58:13-08:00",
   "last_verified": "2026-11-20T00:03:30-08:00",
   "verified_by": "verify.py",
   "status": "verified"
  },
  {
   "id": "ai-incident-database",
   "name": "AI Incident Database",
   "category": "Resources",
   "description": "Searchable public archive of real-world AI incidents.",
   "url": "https://incidentdatabase.ai",
   "lesson": "Look up past failures before you build.",
   "first_seen": "2026-11-19T23:58:13-08:00",
   "last_verified": "2026-11-20T00:03:30-08:00",
   "verified_by": "verify.py",
   "status": "verified"
  }
 ]
}
