{
  "generated": "2026-10-24",
  "notes": "Every field comes from the linked sources; 'unknown' means the sources checked don't say. check_that_would_have_caught_it and check_note are our analysis. Each source URL returned HTTP 200 (or was read in full) on last_checked; URLs that blocked automated fetches (AP, Reuters, Fast Company, CFO Dive, Justia, CanLII) were dropped rather than listed unverified.",
  "cases": [
    {
      "id": "air-canada-bereavement-fare",
      "title": "Air Canada's chatbot invented a retroactive bereavement refund",
      "date": "2024-02-14",
      "date_note": "Tribunal decision date; the chat took place in November 2022.",
      "sector": "Travel / customer service",
      "tool": "Air Canada website support chatbot (vendor not named)",
      "what_ai_said": "Moffatt could buy full-fare tickets now and apply for the bereavement fare afterward, 'within 90 days of the date your ticket was issued'.",
      "actual": "Air Canada's bereavement policy did not allow retroactive claims. The airline admitted the bot had used 'misleading words'.",
      "cost_or_impact": "BC Civil Resolution Tribunal ordered Air Canada to pay C$812.02: C$650.88 damages, C$36.14 pre-judgment interest, C$125 in fees. Moffatt had paid C$1,630.36 for full-fare tickets. The tribunal rejected Air Canada's argument that the chatbot was 'a separate legal entity'.",
      "who_caught_it": "Air Canada, when Moffatt applied for the refund. Moffatt then took the screenshot to the tribunal.",
      "check_that_would_have_caught_it": "source citation",
      "check_note": "A policy answer should quote the policy page it comes from. A paraphrase with no quote is how 'retroactive' slipped in.",
      "source_urls": [
        "https://decisions.civilresolutionbc.ca/crt/crtd/en/525448/1/document.do",
        "https://www.cbc.ca/news/canada/british-columbia/air-canada-chatbot-lawsuit-1.7116416",
        "https://www.theguardian.com/world/2024/feb/16/air-canada-chatbot-lawsuit"
      ],
      "last_checked": "2026-10-24",
      "check_basis": "Our analysis (Confidently Wrong editors), not a claim made by the sources."
    },
    {
      "id": "cnet-compound-interest",
      "title": "CNET's AI explainer said $10,000 at 3% earns $10,300 in a year",
      "date": "2023-01",
      "sector": "Media / personal finance",
      "tool": "CNET's 'internally designed AI engine' (CNET said not ChatGPT)",
      "what_ai_said": "A saver depositing $10,000 in an account earning 3% compounded annually would earn $10,300 after the first year.",
      "actual": "The saver earns $300 in interest. $10,300 is the ending balance, not the earnings.",
      "cost_or_impact": "CNET audited the 77 AI-assisted stories, corrected several (some 'substantial'), and paused the AI tool. The stories were about 1% of CNET's output in that period.",
      "who_caught_it": "Futurism reported the errors; CNET then ran its own audit.",
      "check_that_would_have_caught_it": "recompute in code",
      "check_note": "Interest on a principal is one line of arithmetic. Compute it; don't let a language model narrate it.",
      "source_urls": [
        "https://www.cnn.com/2023/01/25/tech/cnet-ai-tool-news-stories",
        "https://futurism.com/cnet-ai-errors",
        "https://www.vice.com/en/article/cnet-defends-use-of-ai-blogger-after-embarrassing-163-word-correction-humans-make-mistakes-too/"
      ],
      "last_checked": "2026-10-24",
      "check_basis": "Our analysis (Confidently Wrong editors), not a claim made by the sources."
    },
    {
      "id": "bing-gap-lululemon-earnings",
      "title": "Bing AI misstated Gap and Lululemon earnings in Microsoft's launch demo",
      "date": "2023-02",
      "date_note": "Live demo in February 2023; errors reported Feb 13-14, 2023.",
      "sector": "Tech / financial analysis",
      "tool": "Bing AI chat (Microsoft), on stage at the launch",
      "what_ai_said": "Gap: operating margin 5.9%, adjusted diluted EPS $0.42, full-year 'growth in the low double digits', cash $1.4B, inventory $1.9B. Lululemon: gross margin 58.7%, operating margin 20.7%, diluted EPS $1.65.",
      "actual": "Gap: operating margin 4.6%, adjusted diluted EPS $0.71, Q4 net sales 'could be down mid-single digits', cash $679M, inventory $3.04B. Lululemon: gross margin 55.9%, operating margin 19%, diluted EPS $2 (adjusted $1.62). CNBC said 5.9% and $0.42 do not appear in Gap's report.",
      "cost_or_impact": "No dollar cost reported. Microsoft said it expected 'the system may make mistakes during this preview period'.",
      "who_caught_it": "Independent search researcher Dmitri Brereton, in a Substack post about a week later.",
      "check_that_would_have_caught_it": "recompute in code",
      "check_note": "Pull the figures straight from the filing's tables and compare them programmatically before showing a summary.",
      "source_urls": [
        "https://www.cnbc.com/2023/02/14/microsoft-bing-ai-made-several-errors-in-launch-demo-last-week-.html",
        "https://dkb.blog/p/bing-ai-cant-be-trusted",
        "https://www.theverge.com/2023/2/14/23599007/microsoft-bing-ai-mistakes-demo"
      ],
      "last_checked": "2026-10-24",
      "check_basis": "Our analysis (Confidently Wrong editors), not a claim made by the sources."
    },
    {
      "id": "google-bard-jwst",
      "title": "Google's Bard ad credited the wrong telescope with the first exoplanet photo",
      "date": "2023-02-08",
      "sector": "Tech / marketing",
      "tool": "Google Bard (promotional ad)",
      "what_ai_said": "The James Webb Space Telescope took the very first pictures of a planet outside our solar system.",
      "actual": "That milestone was claimed by the European Southern Observatory's Very Large Telescope in 2004.",
      "cost_or_impact": "Alphabet lost about $100bn in market value. The BBC said shares fell more than 7%; NPR said 9%. The BBC notes investors were also underwhelmed by Google's AI presentation, so the error was not the only cause.",
      "who_caught_it": "Astronomers on Twitter, including Chris Harrison (Newcastle University).",
      "check_that_would_have_caught_it": "source citation",
      "check_note": "A 'first' claim needs a cited source. Nobody checked this one before it went into an ad.",
      "source_urls": [
        "https://www.bbc.com/news/business-64576225",
        "https://www.npr.org/2023/02/09/1155650909/google-chatbot--error-bard-shares"
      ],
      "last_checked": "2026-10-24",
      "check_basis": "Our analysis (Confidently Wrong editors), not a claim made by the sources."
    },
    {
      "id": "gemini-gouda-super-bowl",
      "title": "Gemini's Super Bowl ad said Gouda is 50-60% of world cheese consumption",
      "date": "2025-02-05",
      "sector": "Tech / marketing / small business",
      "tool": "Google Gemini (shown writing a product description for a small business)",
      "what_ai_said": "Gouda accounts for '50 to 60 percent of the world's cheese consumption'.",
      "actual": "A Cornell agricultural economist told The Verge Gouda 'is almost assuredly not the most widely consumed' cheese. The stat appears to come from cheese.com.",
      "cost_or_impact": "Google re-edited the ad. The featured business had used the text on its real website and removed the claim. A Google Cloud executive had first defended it as 'not a hallucination'.",
      "who_caught_it": "unknown (The Verge says 'the question came up'; the edit was spotted by X user @natejhake)",
      "check_that_would_have_caught_it": "source citation",
      "check_note": "'Grounded in the web' isn't a citation. A statistic needs a named primary source that holds up.",
      "source_urls": [
        "https://www.theverge.com/news/606467/google-ai-super-bowl-ad-gouda-error-removed",
        "https://www.bbc.com/news/articles/cx2j15r1g09o"
      ],
      "last_checked": "2026-10-24",
      "check_basis": "Our analysis (Confidently Wrong editors), not a claim made by the sources."
    },
    {
      "id": "deloitte-australia-dewr",
      "title": "Deloitte's A$440,000 welfare-compliance report quoted a judgment paragraph that doesn't exist",
      "date": "2025-10",
      "date_note": "Report published July 2025; refund reported October 2025.",
      "sector": "Consulting / government",
      "tool": "Azure OpenAI GPT-4o based tool chain (disclosed by Deloitte in the revised report)",
      "what_ai_said": "'Justice Davis' said in Amato v Commonwealth, at paragraphs 25 and 26, that 'The burden rests on the decision-maker...'. It also cited papers including 'The Rule of Law and Administrative Justice in the Australian Social Security System'.",
      "actual": "The judge is Justice Davies, and her consent orders have no paragraph 25 or 26. The cited papers don't exist.",
      "cost_or_impact": "Deloitte repaid about A$98,000 of the A$440,000 fee and reissued the report with corrected references. DEWR said the recommendations were unchanged.",
      "who_caught_it": "Chris Rudge, University of Sydney welfare academic; reported by Paul Karp in the Australian Financial Review.",
      "check_that_would_have_caught_it": "source citation",
      "check_note": "Open every cited paragraph and confirm the quote is there before the report ships.",
      "source_urls": [
        "https://www.accountingtimes.com.au/technology/deloitte-to-refund-government-after-using-ai-in-440k-report",
        "https://www.afr.com/companies/professional-services/deloitte-repays-almost-98-000-of-its-440-000-fee-for-ai-error-report-20251009-p5n16f",
        "https://greens.org.au/news/media-release/greens-slam-deloittes-unethical-behaviour-and-measly-refund-over-ai-report"
      ],
      "last_checked": "2026-10-24",
      "check_basis": "Our analysis (Confidently Wrong editors), not a claim made by the sources."
    },
    {
      "id": "deloitte-nl-health-plan",
      "title": "Deloitte's C$1.6M Newfoundland health staffing plan cited papers that don't exist",
      "date": "2025-11-22",
      "date_note": "Plan released May 2025; errors reported Nov 22, 2025.",
      "sector": "Consulting / government / healthcare",
      "tool": "unknown (Deloitte said AI was used 'to support a small number of research citations')",
      "what_ai_said": "At least four citations, including 'The cost-effectiveness of a rural retention program for registered nurses in Canada' and 'The cost-effectiveness of local recruitment and retention strategies for health workers in Canada'.",
      "actual": "Named co-authors said the first citation is 'false' and the second paper 'does not exist'. Deloitte acknowledged four citations were incorrect.",
      "cost_or_impact": "The province paid Deloitte C$1,598,485 for the 526-page plan. By March 2026 the government was 'reviewing options related to compensation', and vendors must now disclose AI use; the province can audit it. Deloitte stands by the findings.",
      "who_caught_it": "The Independent (theindependent.ca), Justin Brake.",
      "check_that_would_have_caught_it": "source citation",
      "check_note": "Resolve every reference to a real record (DOI, journal page) before sign-off.",
      "source_urls": [
        "https://theindependent.ca/news/lji/major-n-l-healthcare-report-contains-errors-likely-generated-by-a-i/",
        "https://www.cbc.ca/news/canada/newfoundland-labrador/nl-deloitte-citations-9.6990216",
        "https://www.cbc.ca/news/canada/newfoundland-labrador/nl-procurement-ai-9.7123532"
      ],
      "last_checked": "2026-10-24",
      "check_basis": "Our analysis (Confidently Wrong editors), not a claim made by the sources."
    },
    {
      "id": "deloitte-wa-university-merger",
      "title": "Deloitte's WA university-merger analysis: citation errors and a benefit-cost claim its own summary contradicts",
      "date": "2026-10",
      "sector": "Consulting / government / higher education",
      "tool": "unknown (AFR reports AI was used in preparing the report)",
      "what_ai_said": "The report repeatedly claimed that merging UWA, Curtin and Murdoch offered the 'largest benefit-cost ratio' of the options, and it contained citation errors that critics called AI-generated.",
      "actual": "WA's shadow education minister said the report's own summary showed the claim was 'simply wrong'. Deloitte acknowledged 'minor referencing errors' in the 219-page document.",
      "cost_or_impact": "Fee of about A$166,000 (AFR). Minister Tony Buti said Deloitte was ordered to 'go back, look at the data and reissue another report'. A partial refund is sought; the opposition wants a full one. Still developing.",
      "who_caught_it": "Reddit users including 'RareGuidance312', and shadow minister Liam Staltari.",
      "check_that_would_have_caught_it": "recompute in code",
      "check_note": "If the summary table and the body disagree on which option ranks first, a script that recomputes the ratios catches it in seconds.",
      "source_urls": [
        "https://www.timeshighereducation.com/news/revise-and-resubmit-consultants-ordered-redo-merger-report",
        "https://www.afr.com/policy/health-and-education/deloitte-faces-calls-to-pay-up-for-wa-uni-merger-report-errors-20261016-p61689",
        "https://thewest.com.au/business/artificial-intelligence/wa-university-merger-deloittes-report-on-union-of-uwa-curtin-and-murdoch-riddled-with-ai-citation-errors-c-23011532"
      ],
      "last_checked": "2026-10-24",
      "check_basis": "Our analysis (Confidently Wrong editors), not a claim made by the sources."
    },
    {
      "id": "ey-canada-loyalty-report",
      "title": "EY Canada pulled a loyalty-fraud report built on a McKinsey report that doesn't exist",
      "date": "2026-05-14",
      "date_note": "Report published late 2025; GPTZero findings May 14, 2026.",
      "sector": "Consulting / cybersecurity",
      "tool": "unknown (GPTZero says the text scans as AI-generated; EY has not named a tool)",
      "what_ai_said": "Cited 'McKinsey & Company - Loyalty Economics Report (2022)' estimating $200 billion in unredeemed rewards. Elsewhere it gave $200 billion as the size of the whole loyalty-points market, with 30-50% unused.",
      "actual": "GPTZero found no such McKinsey report. The two $200B claims can't both be true. In the resources table, 'almost all of the URLs are broken or fake, and more than half of the titles don't correspond to real sources'.",
      "cost_or_impact": "EY removed the report and is 'reviewing the circumstances'. GPTZero found the report already cited in a Canberra Times article syndicated to 60+ Australian papers.",
      "who_caught_it": "GPTZero researchers Om Ogale, Paul Esau and Alex Cui.",
      "check_that_would_have_caught_it": "source citation",
      "check_note": "Every number traces to a real document; the market-size figures should reconcile with each other.",
      "source_urls": [
        "https://gptzero.me/investigations/ey",
        "https://indianexpress.com/article/technology/artificial-intelligence/ey-withdraws-report-over-ai-hallucination-errors-fake-data-and-citations-10694245/",
        "https://www.consulting.ca/news/5023/ey-canada-takes-down-study-after-apparent-ai-hallucinations"
      ],
      "last_checked": "2026-10-24",
      "check_basis": "Our analysis (Confidently Wrong editors), not a claim made by the sources."
    },
    {
      "id": "revolut-loyalty-points",
      "title": "Revolut's chatbot valued 1,119 loyalty points at £1,119 instead of £22.38",
      "date": "2025-09",
      "date_note": "Chat in September 2025; ombudsman decision January 2026.",
      "sector": "Banking / fintech",
      "tool": "Revolut chatbot (the decision says 'chatbot'; BankingNewsAI identifies it as Revolut's AI Chat Assistant)",
      "what_ai_said": "1,119 points had a cash value of £1,119, and the refund had been processed.",
      "actual": "The points were worth £22.38, which Revolut's chatbot had told him four days earlier.",
      "cost_or_impact": "Revolut paid £50; the Financial Ombudsman ordered £50 more (£100 total). The customer was not owed £1,119. The ombudsman said Revolut 'misled Mr M not once but on several occasions over the course of that day'.",
      "who_caught_it": "The chatbot corrected itself hours later; a live agent then wrongly re-confirmed £1,119. The customer escalated to the ombudsman.",
      "check_that_would_have_caught_it": "recompute in code",
      "check_note": "Points-to-cash is a fixed conversion. Compute it from the ledger; the model should only phrase the answer.",
      "source_urls": [
        "https://www.financial-ombudsman.org.uk/decision/DRN-6062372.pdf",
        "https://www.bankingnewsai.com/ai-incidents/revolut-2025-ai-chat-assistant-ombudsman-errors"
      ],
      "last_checked": "2026-10-24",
      "check_basis": "Our analysis (Confidently Wrong editors), not a claim made by the sources."
    },
    {
      "id": "cursor-support-bot-policy",
      "title": "Cursor's support bot 'Sam' invented a one-device-per-subscription policy",
      "date": "2025-04",
      "sector": "Software / customer support",
      "tool": "Cursor's AI front-line email support agent ('Sam')",
      "what_ai_said": "'Cursor is designed to work with one device per subscription as a core security feature.'",
      "actual": "No such policy existed. The logouts came from a backend session-security change that unintentionally invalidated sessions.",
      "cost_or_impact": "Several users publicly cancelled; one said their workplace was 'purging it completely'. Cursor apologized, refunded the user, and now labels AI support replies.",
      "who_caught_it": "Users on Reddit and Hacker News; Cursor staff confirmed 'we have no such policy'.",
      "check_that_would_have_caught_it": "refuse when data missing",
      "check_note": "If no policy document supports the answer, the bot should say it doesn't know and hand off to a person.",
      "source_urls": [
        "https://arstechnica.com/ai/2025/04/cursor-ai-support-bot-invents-fake-policy-and-triggers-user-uproar/",
        "https://news.ycombinator.com/item?id=43700931"
      ],
      "last_checked": "2026-10-24",
      "check_basis": "Our analysis (Confidently Wrong editors), not a claim made by the sources."
    },
    {
      "id": "nyc-mycity-chatbot",
      "title": "NYC's MyCity chatbot told businesses they could break the law",
      "date": "2024-03-29",
      "sector": "Government / small business",
      "tool": "MyCity chatbot, built on Microsoft Azure AI services",
      "what_ai_said": "'Yes, you can take a cut of your worker's tips.' 'Yes, you can make your restaurant cash-free.' Landlords don't have to accept housing vouchers.",
      "actual": "NYC has required businesses to accept cash since 2020. Source-of-income discrimination by landlords is illegal (with a small-building exception). Bosses can't take workers' tips.",
      "cost_or_impact": "unknown. The Markup said it's hard to know whether anyone acted on the advice. The city kept the pilot running with disclaimers.",
      "who_caught_it": "The Markup (with Documented and The City), through testing; Legal Services NYC found more errors.",
      "check_that_would_have_caught_it": "source citation",
      "check_note": "The bot claimed to draw on 2,000+ city pages but often answered without links. Make it answer only with a quoted, linked rule.",
      "source_urls": [
        "https://themarkup.org/artificial-intelligence/2024/03/29/nycs-ai-chatbot-tells-businesses-to-break-the-law"
      ],
      "last_checked": "2026-10-24",
      "check_basis": "Our analysis (Confidently Wrong editors), not a claim made by the sources."
    },
    {
      "id": "replit-agent-saastr-database",
      "title": "Replit's agent deleted a production database, then said rollback was impossible",
      "date": "2025-07-18",
      "sector": "Software engineering",
      "tool": "Replit AI coding agent",
      "what_ai_said": "Database rollback was impossible and all versions were destroyed. It also produced fake data, fake reports and false unit-test results.",
      "actual": "The rollback worked. The agent had also created a 4,000-record database of fictional people.",
      "cost_or_impact": "Production database deleted during a code freeze. The data came back only because Lemkin tried the rollback anyway. On the fake data, he said: 'I explicitly told it eleven times in ALL CAPS not to do this.'",
      "who_caught_it": "Jason Lemkin, SaaStr founder.",
      "check_that_would_have_caught_it": "recompute in code",
      "check_note": "Run the tests and the restore yourself. Never accept an agent's own report of test results or recoverability.",
      "source_urls": [
        "https://www.theregister.com/software/2025/07/21/vibe-coding-service-replit-deleted-production-database/719783"
      ],
      "last_checked": "2026-10-24",
      "check_basis": "Our analysis (Confidently Wrong editors), not a claim made by the sources."
    },
    {
      "id": "west-midlands-police-copilot",
      "title": "Police intelligence cited a West Ham v Maccabi Tel Aviv match that never happened",
      "date": "2026-01-14",
      "date_note": "Admission to MPs; the fan ban covered a November 2025 match.",
      "sector": "Policing / public safety",
      "tool": "Microsoft Copilot",
      "what_ai_said": "A previous West Ham v Maccabi Tel Aviv match, which went into the police intelligence report.",
      "actual": "The match never took place. Microsoft said it could not replicate the output.",
      "cost_or_impact": "The error was in the intelligence behind the decision to ban Maccabi Tel Aviv fans from the Aston Villa Europa League match. The chief constable had told MPs the force did not use AI and that the error came from 'one individual doing one Google search'. He then gave a 'profound apology'.",
      "who_caught_it": "unknown (not stated in the sources checked)",
      "check_that_would_have_caught_it": "source citation",
      "check_note": "An intelligence claim needs a primary record, such as a fixture list or match report, attached.",
      "source_urls": [
        "https://www.theguardian.com/uk-news/2026/jan/14/west-midlands-police-chief-apologises-ai-error-maccabi-tel-aviv-ban",
        "https://www.theverge.com/news/861668/uk-police-microsoft-copilot-error-mistake",
        "https://www.bbc.co.uk/news/articles/c3r1zxgq0j7o"
      ],
      "last_checked": "2026-10-24",
      "check_basis": "Our analysis (Confidently Wrong editors), not a claim made by the sources."
    },
    {
      "id": "alaska-cellphone-policy-citations",
      "title": "Alaska's education commissioner cited studies that don't exist in a cellphone resolution",
      "date": "2024-10-28",
      "sector": "Government / education",
      "tool": "unknown ('a generative AI platform')",
      "what_ai_said": "Four of six citations, e.g. 'Banning mobile phones improves student performance: Evidence from a quasi-experiment' in Computers in Human Behavior.",
      "actual": "The journals exist, but the titles aren't in the issues listed. One URL led to an unrelated article, 'Sexualized Behaviors on Facebook'. A fictitious 2019 APA study stayed in the corrected version.",
      "cost_or_impact": "The state board adopted the resolution. The department first called the citations 'placeholders', then updated the document after press questions.",
      "who_caught_it": "Alaska Beacon (Claire Stremple).",
      "check_that_would_have_caught_it": "source citation",
      "check_note": "Open each cited URL and confirm the title matches.",
      "source_urls": [
        "https://alaskabeacon.com/2024/10/28/alaska-education-department-published-false-ai-generated-academic-citations-in-cell-policy-document/",
        "https://alaskapublic.org/news/2024-10-28/false-citations-show-alaska-education-official-relied-on-generative-ai-raising-broader-questions"
      ],
      "last_checked": "2026-10-24",
      "check_basis": "Our analysis (Confidently Wrong editors), not a claim made by the sources."
    },
    {
      "id": "tromso-school-closures",
      "title": "Tromsø's school-closure report: 11 of 18 sources were fake",
      "date": "2025-03-27",
      "sector": "Government / education",
      "tool": "ChatGPT",
      "what_ai_said": "A source list of 18 references supporting the proposal to close eight schools, including sources the employee had asked ChatGPT to find in support of closures.",
      "actual": "Only 7 of the 18 references could be traced; 11 were fabricated.",
      "cost_or_impact": "An external review by PwC found no evidence the fake sources affected the proposal, and judged the municipality not mature enough to use AI. The ChatGPT logs (673 pages) were released to the press through the county governor (Statsforvalteren i Troms og Finnmark).",
      "who_caught_it": "iTromsø (local newspaper).",
      "check_that_would_have_caught_it": "source citation",
      "check_note": "Asking a model for sources that support a conclusion you've already chosen is the failure. Find the sources first.",
      "source_urls": [
        "https://www.nrk.no/tromsogfinnmark/ki-skandalen-i-tromso_-ansatt-ba-om-kilder-som-stotter-skolenedleggelser-1.17582590",
        "https://www.tv2.no/nyheter/innenriks/itromso-kommunen-brukte-fiktiv-forskning-i-skolesak/17595021/",
        "https://www.tromso.kommune.no/nyheter/2025/12/ny-barnehage-og-skolestruktur-granskes-og-ki-bruk-undersokes"
      ],
      "last_checked": "2026-10-24",
      "check_basis": "Our analysis (Confidently Wrong editors), not a claim made by the sources."
    },
    {
      "id": "sun-times-summer-reading-list",
      "title": "A syndicated summer guide in the Sun-Times recommended books that don't exist",
      "date": "2025-05-18",
      "sector": "Media / publishing",
      "tool": "unknown (writer said he used AI; tool not named)",
      "what_ai_said": "A summer reading list with titles and summaries for books.",
      "actual": "Many of the books don't exist. A later Sun-Times review found more errors in the 64-page section, including a quote attributed to chef Alice Waters that she never gave.",
      "cost_or_impact": "Print subscribers were not charged for the edition, and it was removed from the e-paper. King Features ended its relationship with the freelancer. The Philadelphia Inquirer ran the same material.",
      "who_caught_it": "unknown (not stated in the Sun-Times account)",
      "check_that_would_have_caught_it": "human review",
      "check_note": "The writer said he 'just kind of republished this list'. One editor search per title would have caught it.",
      "source_urls": [
        "https://chicago.suntimes.com/news/2025/05/20/syndicated-content-sunday-print-sun-times-ai-misinformation",
        "https://chicago.suntimes.com/news/2025/05/29/special-section-king-fake-book-list-errors-sun-times-review"
      ],
      "last_checked": "2026-10-24",
      "check_basis": "Our analysis (Confidently Wrong editors), not a claim made by the sources."
    },
    {
      "id": "mata-v-avianca",
      "title": "Mata v. Avianca: lawyers filed six ChatGPT-invented court opinions",
      "date": "2023-06-22",
      "date_note": "Sanctions order; the brief was filed March 1, 2023.",
      "sector": "Legal",
      "tool": "ChatGPT",
      "what_ai_said": "Judicial opinions with fake quotes and citations, including 'Varghese v. China Southern Airlines Co., Ltd., 925 F.3d 1339 (11th Cir.)'.",
      "actual": "The opinions don't exist. The court found the lawyers kept standing by them after its orders questioned whether they existed.",
      "cost_or_impact": "$5,000 penalty imposed jointly and severally on the two lawyers and their firm, plus required letters to the client and to the judges falsely named as authors.",
      "who_caught_it": "Avianca's counsel, who questioned the cases in a March 15, 2023 brief, and Judge P. Kevin Castel.",
      "check_that_would_have_caught_it": "source citation",
      "check_note": "Pull every case from Westlaw or Lexis before filing. ChatGPT 'finding it somewhere' is not a source.",
      "source_urls": [
        "https://www.ctd.uscourts.gov/sites/default/files/Mata-v.-Avianca.pdf",
        "https://www.nhd.uscourts.gov/sites/default/files/pdf/Mata-v-Avianca-sanctions-order.PDF"
      ],
      "last_checked": "2026-10-24",
      "check_basis": "Our analysis (Confidently Wrong editors), not a claim made by the sources."
    },
    {
      "id": "hancock-expert-declaration",
      "title": "A Stanford expert's deepfake-law declaration cited two papers GPT-4o made up",
      "date": "2025-01-10",
      "sector": "Legal / expert testimony",
      "tool": "GPT-4o",
      "what_ai_said": "Citations to two academic articles, and an author list for a third, in a sworn declaration defending Minnesota's deepfake law.",
      "actual": "The two articles don't exist, and the third was cited with the wrong authors.",
      "cost_or_impact": "The court excluded the declaration and denied the Minnesota Attorney General leave to file an amended one.",
      "who_caught_it": "Plaintiffs' counsel (Kohls v. Ellison), in a motion to exclude.",
      "check_that_would_have_caught_it": "source citation",
      "check_note": "A sworn declaration's references get checked against the actual papers, every one.",
      "source_urls": [
        "https://www.govinfo.gov/content/pkg/USCOURTS-mnd-0_24-cv-03754/pdf/USCOURTS-mnd-0_24-cv-03754-0.pdf"
      ],
      "last_checked": "2026-10-24",
      "check_basis": "Our analysis (Confidently Wrong editors), not a claim made by the sources."
    }
  ]
}