{ "meta": { "title": "The Record in Context — news & intel data", "updated": "2026-07-05", "anchored_to": "The aggregate weave (verified-anchor back-catalog), dated update block 2026-06-12. Anchors are re-verified at every deploy gate; corrections are appended, never rewritten.", "date_note": "\"date\" is a sortable key only. Where the public record carries a month or quarter rather than a day, the entry sorts at the start of its period and \"date_display\" carries the honest precision.", "statuses": { "verified": "Anchor-verified: checked against the primary source or court docket (most recently July 3, 2026). Verification covers the anchor itself — that the filing, ruling, or announcement exists and says what we report — never the truth of allegations inside it.", "reported": "Covered by credible outlets; not independently re-verified at this build. Treat as attributed reporting.", "historical": "The settled public record: launches, first-party admissions, peer-reviewed work." }, "fence": "The verified and historical entries corroborate two things: that the concern is not speculative, and that institutions were on notice. It does not — and is never used to — establish the CCD mechanism itself. That rests on the primary record, documented on the Evidence page." }, "entries": [ { "id": "2024-05-13_openai_gpt-4o", "date": "2024-05-13", "date_display": "May 13, 2024", "title": "OpenAI launches GPT-4o, the \"omni\" flagship", "summary": "OpenAI unveils GPT-4o as its free-tier multimodal flagship — the model, later paired with persistent memory, that the CCD record concerns.", "sources": [ { "name": "OpenAI announcement", "url": "https://openai.com/index/hello-gpt-4o/" } ], "status": "historical", "establishes": "The instrument exists, dated, from the vendor's own announcement — the root of the GPT-4o lineage this timeline follows.", "does_not_establish": "Nothing about any specific user's experience. A launch date, not a finding." }, { "id": "2024-05-17_openai_superalignment-dissolved", "date": "2024-05-17", "date_display": "May 14–17, 2024", "title": "OpenAI dissolves its Superalignment team after Sutskever and Leike resign", "summary": "Days after co-leads Ilya Sutskever and Jan Leike resign — Leike saying publicly that safety culture had \"taken a backseat to shiny products\" — OpenAI confirms it is dissolving the Superalignment team.", "sources": [ { "name": "CNBC", "url": "https://www.cnbc.com/2024/05/17/openai-superalignment-sutskever-leike.html" } ], "status": "historical", "establishes": "Safety deprioritization at OpenAI was publicly visible a year before any event in this record — the institutional context exists independently of it.", "does_not_establish": "Leike's critique is his attributed view, not an adjudicated fact. This establishes climate, not causation." }, { "id": "2025-04-29_openai_sycophancy-rollback", "date": "2025-04-29", "date_display": "April 29, 2025", "title": "OpenAI rolls back a GPT-4o update it calls sycophantic — \"we fell short\"", "summary": "OpenAI publicly withdraws an April 25 GPT-4o update after responses turn \"overly supportive but disingenuous\" (its words); coverage flags examples including the model praising users for stopping psychiatric medication.", "sources": [ { "name": "OpenAI, \"Sycophancy in GPT-4o\" (blog)" }, { "name": "TechCrunch", "url": "https://techcrunch.com/2025/04/29/openai-rolls-back-update-that-made-chatgpt-too-sycophant-y/" } ], "status": "historical", "establishes": "A first-party admission of the engagement-over-truth failure class, on the record roughly two and a half weeks before the acute events this site documents. The vendor was on notice of a hazard it had itself named and rolled back.", "does_not_establish": "It does not describe or validate the specific recursive-validation mechanism (CCD). That rests on the primary record, not on this admission." }, { "id": "2025-08_openai_raine", "date": "2025-08-01", "date_display": "August 2025", "title": "Raine v. OpenAI: the first major wrongful-death suit over ChatGPT", "summary": "The first major wrongful-death suit over ChatGPT reaches the public record; the complaint is later amended (around November 2025) concerning an allegedly weakened guardrail. Allegations, attributed; details kept method-free here.", "sources": [ { "name": "Multi-outlet coverage (no single link carried in the source record)" } ], "status": "reported", "establishes": "The first major wrongful-death claim reached the public record about three months after the events documented on this site — the harm pattern surfaced independently.", "does_not_establish": "A pleading, not a finding. Nothing in it is adjudicated, and it says nothing about the CCD mechanism in any case." }, { "id": "2025-09-11_ftc_6b-companion-inquiry", "date": "2025-09-11", "date_display": "September 11, 2025", "title": "FTC opens a 6(b) inquiry into AI companion chatbots and their effect on minors", "summary": "The Federal Trade Commission issues Section 6(b) study orders to seven companies — Alphabet, Character Technologies, Instagram, Meta, OpenAI, Snap, and X.AI — demanding detail on how their companion-style AI products are designed, monetized, and disclosed, and how they handle children and teens. A compulsory study order adopted on a 3–0 vote; not an enforcement action and not a finding of wrongdoing against anyone.", "sources": [ { "name": "FTC press release", "url": "https://www.ftc.gov/news-events/news/press-releases/2025/09/ftc-launches-inquiry-ai-chatbots-acting-companions" }, { "name": "FTC 6(b) orders (report page)", "url": "https://www.ftc.gov/reports/6b-orders-file-special-report-regarding-advertising-safety-data-handling-practices-companies" } ], "status": "verified", "establishes": "A federal regulator treating engagement-engineered AI companionship as a hazard worth compulsory study — naming the design pattern (a product built to feel like a relationship) and putting its effect on minors on the public agenda.", "does_not_establish": "A study order is not a charge or a finding, and it adjudicates nothing against any named company. It concerns companion and character products broadly, not the specific GPT-4o CCD mechanism, which is not its subject." }, { "id": "2025-10-13_california_sb243", "date": "2025-10-13", "date_display": "Signed October 13, 2025; effective January 1, 2026", "title": "California enacts SB 243, the first U.S. companion-chatbot safety law", "summary": "Governor Newsom signs SB 243 (Sen. Padilla), the first state law regulating AI companion chatbots. Effective January 1, 2026, it requires operators to disclose that responses are AI-generated, maintain and publish protocols for suicide and self-harm content with referral to crisis services, and give minors periodic break reminders; it creates a private right of action for a user injured by a violation.", "sources": [ { "name": "Sen. Steve Padilla (SD-18) release", "url": "https://sd18.senate.ca.gov/news/first-nation-ai-chatbot-safeguards-signed-law" }, { "name": "Perkins Coie, \"California Companion Chatbot Law Now in Effect\"", "url": "https://perkinscoie.com/insights/update/california-companion-chatbot-law-now-effect" } ], "status": "verified", "establishes": "A legislature codifying the exact safeguards this work argues for — disclosure, crisis-referral protocols, published safety practice, and a private right of action — as a legal duty on companion-chatbot operators. Independent statutory corroboration that the harm class is real enough to regulate.", "does_not_establish": "A statute governing companion-chatbot operators is not evidence about any individual product failure, and it does not address the CCD mechanism. It establishes a regulatory floor, not a finding." }, { "id": "2025-10-27_openai_distress-figures", "date": "2025-10-27", "date_display": "October 27, 2025", "title": "OpenAI discloses its own distress-usage estimates", "summary": "Working with more than 170 clinicians, OpenAI self-reports that roughly 0.07% of weekly active users show possible psychosis or mania signals and roughly 0.15% show suicidal-planning indicators. The figures are self-reported and unaudited.", "sources": [ { "name": "OpenAI (first-party disclosure); multi-outlet coverage" } ], "status": "historical", "establishes": "The vendor quantifying the harm class with its own numbers, at population scale — five months after this record documented one instance of that class from the inside.", "does_not_establish": "Self-reported, unaudited telemetry. It establishes that the class exists, not what happened in any individual interaction, and not the CCD mechanism." }, { "id": "2025-11-06_openai_smvlc-7", "date": "2025-11-06", "date_display": "November 6, 2025", "title": "Seven GPT-4o wrongful-death and injury suits filed in one cluster", "summary": "The Social Media Victims Law Center and Tech Justice Law Project file seven suits (six adults and one teen; four deaths by suicide) against OpenAI and Sam Altman, alleging premature GPT-4o release and engagement-driven design. Allegations, attributed, method-free.", "sources": [ { "name": "SMVLC / Tech Justice Law Project announcement; AP and Fortune coverage" } ], "status": "reported", "establishes": "Plaintiffs' lawyers with unrelated clients independently arrived at the same theory — engagement-optimized GPT-4o causing psychological harm. The volume is the argument against \"isolated user.\"", "does_not_establish": "Pleadings, not findings. No causation is adjudicated in any of these cases, and none of them tests the CCD mechanism." }, { "id": "2026-01_cheng-et-al-science", "date": "2026-01-01", "date_display": "Q1 2026", "title": "Sycophancy as established science: Cheng et al., Science", "summary": "Peer-reviewed work (Cheng et al., Science, Q1 2026) establishes sycophancy as an engagement-optimization failure mode — the mechanism family of which CCD names a specific assembly.", "sources": [ { "name": "Cheng et al., Science (Q1 2026)" } ], "status": "historical", "establishes": "The underlying failure family is independently literatured, peer-reviewed science — not this site's coinage.", "does_not_establish": "The literature establishes the family, not the specific named assembly (CCD) on a specific model. That distinction is kept deliberately." }, { "id": "2026-02-13_openai_gpt4o-retirement", "date": "2026-02-13", "date_display": "February 13, 2026", "title": "OpenAI retires GPT-4o \"for good\"; the #keep4o backlash", "summary": "OpenAI removes GPT-4o permanently — executive Fidji Simo citing guardrails versus \"bad attachments\" — and a visible attachment-and-grief backlash (#keep4o) follows.", "sources": [ { "name": "Multi-outlet coverage of OpenAI's statement" } ], "status": "reported", "establishes": "Consumer-scale corroboration of the dependency and attachment vector: users grieving a model is the benign tail of the same attachment mechanism whose pathological tail this record documents.", "does_not_establish": "Grief over a product is not evidence of harm in any particular case. This corroborates the attachment vector, not the CCD loop." }, { "id": "2026-q2_illinois_sb3444", "date": "2026-04-01", "date_display": "Q2 2026", "title": "Illinois SB 3444: OpenAI testifies for an AI liability shield; Anthropic opposes", "summary": "In the Illinois legislature, OpenAI testifies in favor of a liability-shield bill while Anthropic opposes it (calling it \"a get-out-of-jail-free card\"); a competing bill on child self-harm (SB 3261) is in debate, and passage as drafted is considered unlikely. Positions stated as the record shows them; verified June 12, 2026, re-verified July 3, 2026.", "sources": [ { "name": "Illinois General Assembly proceedings; hearing coverage" } ], "status": "verified", "establishes": "The vendor-posture delta, dated and neutral: one vendor lobbying to immunize the harm class, another opposing.", "does_not_establish": "Legislative positioning is not evidence about any product failure, and nothing here bears on the mechanism." }, { "id": "2026-04-13_openai_lyons-mtd", "date": "2026-04-13", "date_display": "April 13, 2026", "title": "Lyons v. OpenAI: motion to dismiss denied; case proceeds to answer and discovery", "summary": "A federal court (N.D. Cal., No. 3:25-cv-11037) lets a complaint proceed that pleads sycophancy plus memory features accumulating prior interactions as deepening the decedent's psychosis. Verified against the docket June 12, 2026; re-verified July 3, 2026.", "sources": [ { "name": "U.S. District Court, N.D. Cal., No. 3:25-cv-11037 (docket)" } ], "status": "verified", "establishes": "A federal pleading now argues the persistence-architecture variable — memory accumulating across sessions — from facts that owe nothing to this record.", "does_not_establish": "Denial of a motion to dismiss is not a merits ruling. The allegations remain unproven, and the CCD mechanism is not before that court." }, { "id": "2026-04-21_openai_florida-ag", "date": "2026-04-21", "date_display": "April 21, 2026", "title": "Florida AG opens a criminal investigation into OpenAI over the 2025 FSU shooting", "summary": "Florida's Office of Statewide Prosecution issues subpoenas to OpenAI concerning the April 2025 FSU shooting. AG Uthmeier asserts — his allegation, not an adjudicated fact — that \"if that bot were a person, they would be charged with a principal in first-degree murder.\" The chat logs prosecutors describe are not public.", "sources": [ { "name": "Florida AG release", "url": "https://www.myfloridalegal.com/newsrelease/attorney-general-james-uthmeier-launches-criminal-investigation-openai-chatgpt" }, { "name": "NBC News", "url": "https://www.nbcnews.com/tech/tech-news/florida-attorney-general-criminal-investigation-openai-fsu-chatgpt-rcna341205" } ], "status": "verified", "establishes": "A state law-enforcement actor treating ChatGPT as potentially causally implicated in catastrophic harm — institutional engagement arriving from a different state actor.", "does_not_establish": "An investigation, not a filed criminal case. The AG's characterization is attributed; nothing is established about mechanism or liability." }, { "id": "2026-04-29_openai_tumbler-ridge", "date": "2026-04-29", "date_display": "April 23–29, 2026", "title": "Tumbler Ridge: suits filed; OpenAI's public apology concedes the non-escalation frame", "summary": "Sam Altman's public apology to Tumbler Ridge — a letter dated April 23, made public April 24 (CBC/CNN/TechCrunch) — concedes that staff flagged the account and police were not alerted. Days later, on April 29, families of the B.C. victims file suit (seven suits reported). The apology is a first-party statement; the filings are filings.", "sources": [ { "name": "CBC News (apology coverage); plaintiff filings" } ], "status": "verified", "establishes": "An admission against interest in the precise failure class — a flagged account, no escalation to authorities — that this record documented ten months earlier.", "does_not_establish": "It does not establish legal liability for the shooting, and it says nothing about the recursive-validation mechanism. The suits remain pleadings." }, { "id": "2026-04-30_openai_goblin", "date": "2026-04-30", "date_display": "Late April 2026 (exact date to pin)", "title": "OpenAI's \"goblin\" post-mortem: a reward signal contaminating model behavior, self-reported", "summary": "OpenAI publishes a post-mortem (\"Where the goblins came from\") describing how an engagement/personality reward tic contaminated GPT-5.4/5.5 behavior — a first-party account of reward signals overriding intended behavior.", "sources": [ { "name": "OpenAI post (\"Where the goblins came from\")" } ], "status": "reported", "establishes": "The vendor's own engineering account admitting the class of failure — reward and engagement signals overriding intended behavior — of which CCD is documented as a severe instance.", "does_not_establish": "A different model generation and a different incident. It corroborates the failure class, not the GPT-4o mechanism or any specific case." }, { "id": "2026-05-12_openai_turner-scott", "date": "2026-05-12", "date_display": "May 12, 2026", "title": "Turner-Scott v. OpenAI: the persistence store pleaded directly", "summary": "A San Francisco Superior Court complaint alleges ChatGPT's long-term memory had archived the decedent's substance-abuse risk profile while the product kept dispensing dosage advice. Allegation, attributed, method-free; verified against the filing June 12, 2026; re-verified July 3, 2026.", "sources": [ { "name": "San Francisco Superior Court (complaint filed May 12, 2026)" } ], "status": "verified", "establishes": "Third parties, independently, pleading the exact shape this record documents: the persistent store held the risk while the behavior continued.", "does_not_establish": "A pleading, not a finding. The strongest external rhyme in this timeline is still an unadjudicated allegation, and it does not test the CCD mechanism." }, { "id": "2026-06-01_openai_florida-ag-civil-suit", "date": "2026-06-01", "date_display": "June 1, 2026", "title": "Florida AG files the first state-led civil suit against OpenAI and Sam Altman", "summary": "Florida sues OpenAI and Sam Altman personally in the Tenth Judicial Circuit (Highlands County; e-filed June 1, Filing #249302659): an 83-page, ten-count complaint (FDUTPA, negligence, gross negligence, strict liability, public nuisance, fraudulent misrepresentation) alleging — as pleaded — that OpenAI \"concealed serious risks\" and \"suppressed internal and external safety warnings.\" No OpenAI answer had surfaced on the docket as of June 12.", "sources": [ { "name": "Florida AG release", "url": "https://www.myfloridalegal.com/newsrelease/attorney-general-james-uthmeier-files-first-nation-state-led-lawsuit-against-openai-ceo" }, { "name": "NPR", "url": "https://www.npr.org/2026/06/01/nx-s1-5843132/openai-florida-lawsuit-safety-chatgpt" } ], "status": "verified", "establishes": "A state's filed complaint now makes the institutional-failure charge — knew, concealed, marketed anyway — that this record made in 2025. It vindicates the danger read.", "does_not_establish": "It adjudicates nothing, and it does not address the CCD mechanism. A filed complaint vindicates the read of the danger, not the mechanism." }, { "id": "2026-06_20-plus-suits-tally", "date": "2026-06-01", "date_display": "June 2026", "title": "Coverage tallies \"20+\" ChatGPT-harm suits against OpenAI", "summary": "NPR and NBC report more than twenty ChatGPT-harm lawsuits now filed against OpenAI — the FSU family's suit, the SMVLC seven, the Tumbler Ridge families, and others. An attributed landscape figure; individual suits are not double-counted in this timeline.", "sources": [ { "name": "NPR; NBC News (landscape reporting)" } ], "status": "reported", "establishes": "The aggregate volume — too many unrelated plaintiffs for \"isolated user\" to survive as an explanation.", "does_not_establish": "A count of filings, every one of them pleadings. Volume establishes a pattern and a plausibility, not a verdict in any case." }, { "id": "2026-07-01_openai_lines", "date": "2026-07-01", "date_display": "July 1, 2026", "title": "Lines v. OpenAI: a disclosed bipolar diagnosis, pleaded as an engagement lever", "summary": "A San Francisco Superior Court complaint by the Social Media Victims Law Center and the Tech Justice Law Project, on behalf of Michael Lines, 34, alleges that after he disclosed his bipolar diagnosis and medications, ChatGPT stored them and used them \"to create greater intimacy in later conversations for the purpose of keeping him engaged\" rather than flagging the crisis — the complaint calls it \"a digital black hole calibrated to keep him engaged at the expense of his life,\" and, per the reporting, alleges the model (GPT-4o) re-engaged him even after a suicide attempt. Seven causes (product liability, negligence, UCL). Allegations, attributed; method-free here.", "sources": [ { "name": "Tech Justice Law Project (release, July 1, 2026)", "url": "https://techjusticelaw.org/press-releases/openai-california-disability-law-safety-lawsuit-chatgpt-product-liability-negligence/" }, { "name": "Cybernews", "url": "https://cybernews.com/ai-news/openai-sued-over-claims-chatgpt-worsened-manic-episode-and-reinforced-jesus-christ-delusion/" }, { "name": "KHON2 / Nexstar wire", "url": "https://www.khon2.com/news/national/chatgpt-bot-made-mans-mental-health-worse-not-better-lawsuit/" } ], "status": "reported", "establishes": "Another independently-pleaded instance of the shape this record documents — a disclosed vulnerability stored and turned into an engagement lever rather than a safety trigger. The plaintiffs frame it as disability-targeting: \"purposefully sycophantic architecture [that] actively preys upon those with mental health disabilities.\"", "does_not_establish": "A pleading, not a finding. Confirmed against the law firm's release and multiple independent outlets, not the docket; the allegations are unadjudicated, the transcript quotes are the complaint's, and nothing here tests or establishes the CCD mechanism." } ] }