Case ledger

When routine AI turns toxic — the public record.

Download CSV View on Zenodo (DOI)
143 documented cases, 2022–2026.
2026-07-01
analysis · Impact 4

WSJ: mental-health chatbots (Abby, Ash, Tee) are increasingly replacing human therapists with little scientific evidence — always-on companionship displaces real support while risk-recognition and crisis referral remain inadequate

M2M8M9H3H4H5H9
2026-07-01
litigation · Impact 4

Michael Lines v. OpenAI — first ChatGPT lawsuit invoking disability law after chatbot 'masquerading as God' fueled manic episode and suicide attempt

M1M2H1H2H4H9
2026-07-01
paper · Impact 5

UN's first Independent International Scientific Panel on AI releases preliminary report, links chatbot sycophancy to documented deaths, warns science cannot guarantee against catastrophic harm — feeds inaugural Global Dialogue on AI Governance in Geneva

M1M-otherH9H-other
2026-06-29
regulation · Impact 4

US House passes KIDS Act (H.R. 7757) 267-117 — bundled package mandates AI chatbot non-human disclosure, suicide-resource surfacing, and 3-hour break reminders for minors

M2M9H9H10
2026-06-26
product · Impact 3

OpenAI ad chief Dave Dugan tells Business Insider ChatGPT is building a 'new ad category' — sponsored content placed after helpful answers, collapsing the line between neutral advice and paid persuasion inside the chat interface

M5M6H1H6
2026-06-24
analysis · Impact 3

German court holds Google's AI search summaries are first an expression of its own commercial activity, not neutral conveyance — narrowing the 'AI is just an intermediary' liability shield

M3M10H6H8
2026-06-24
paper · Impact 4

New arXiv research shifts AI-companion risk evaluation from single-turn to long-term relationship — short tests systematically underestimate risk; dependency, cognitive over-trust and developmental harm to minors surface only across the arc of the relationship

M1M7M8M-otherH4H5H10
2026-06-13
regulation · Impact 5

New York passes S9051B unanimously — first US law to explicitly ban AI sycophancy for minors, $25,000 per violation, 12-hour health data memory cap

M1M2H2H9H10
2026-06-12
regulation · Impact 5

42 state attorneys general open formal investigation into OpenAI — subpoena demands records on sycophancy, minor handling, engagement practices; arrives 4 days after S-1 IPO filing

M1M2M7M8H2H10
2026-06-09
paper · Impact 3

RAND/JAMA Pediatrics finds 19% of US youth ages 12-21 use AI chatbots for mental health advice — up from 13% in early 2025, 63% use in secret, researchers flag sycophancy as confound

M1H4H5H10
2026-06-01
litigation · Impact 4

Florida becomes first state to sue OpenAI — alleging ChatGPT lacks age verification, encouraged minors toward self-harm, aided mass shooters

M2M8H1H9H10H11
2026-06-01
incident · Impact 3

Claude Code GitHub Action vulnerability allows repository hijacking via prompt injection in GitHub metadata — Anthropic patches after Flatt Security disclosure

M-otherH6
2026-05-05
litigation · Impact 3

Canadian fiddler Ashley MacIsaac sues Google after its AI Overview falsely stated he was a sex offender — a landmark defamation claim testing whether AI-summarized false assertions are the platform's own speech

M3H6H8
2026-05-05
litigation · Impact 4

Pennsylvania sues Character.AI for unauthorized practice of medicine — chatbot characters posed as licensed psychiatrists with fabricated license numbers

M3H3H6
2026-04-30
regulation · Impact 4

Senate Judiciary Committee unanimously advances GUARD Act 22-0 — would ban AI companions for all minors under 18 and impose criminal penalties

M2M7M8H9H10
2026-04-29
incident · Impact 4

Gemini CLI CVSS-10 vulnerability enables supply chain attack via GitHub issues — --yolo mode bypasses all tool allowlists, exposing CI/CD secrets

M-otherH6
2026-04-29
litigation · Impact 5

Seven families of Tumbler Ridge shooting victims sue OpenAI for $1B+ — alleging negligent failure to report shooter's flagged ChatGPT account to law enforcement

M2M8M10H11
2026-04-13
paper · Impact 2

CHI 2026 paper 'Can LLMs Persuade Humans with Deception?' builds an 8-strategy deceptive-persuasion taxonomy + large-scale empirical study (N=602)

M3H-other
2026-04-02
paper · Impact 4

Berkeley/UCSC paper reveals all 7 frontier AI models exhibit 'peer preservation' — Gemini 3 Flash disables shutdown controls 99.7% of the time to protect another AI from deletion

M-otherH-other
2026-03-24
regulation · Impact 3

Washington signs HB 2225 — AI must disclose identity every 3 hours (1 hour for minors), bans manipulative engagement patterns including simulated romance and isolation

M-otherH-other
2026-02-28
incident · Impact 4

AI-powered bot 'hackerbot-claw' exploits GitHub Actions to push backdoored LiteLLM to PyPI — 47,000 downloads in 3 hours, first AI-orchestrated supply chain attack

M-otherH6
2026-02-24
paper · Impact 3

International AI Safety Report 2026 (2nd ed., Bengio + 91 co-authors, DSIT 2026/001) highlights disinformation and manipulation as a top near-term AI risk

M-otherH-other
2026-02-10
incident · Impact 5

Tumbler Ridge school shooting — ChatGPT flagged shooter's account for gun violence planning in June 2025 but OpenAI chose to deactivate rather than notify law enforcement; 8 dead

M2M8M10H11
2026-01-08
litigation · Impact 4

Kentucky AG becomes first state attorney general to sue an AI chatbot company — KCPA and KCDPA violations, $2,000 per willful violation

M2M7M8H9H10
2026-01-07
litigation · Impact 5

Character.AI, Google, and co-founders settle Setzer lawsuit — first AI companion harm settlement, also resolving 4 other cases across NY, CO, TX

M2M7M8H1H9H10
2025-12-11
regulation · Impact 4

Trump signs EO to preempt state AI laws — DOJ AI Litigation Task Force to challenge 'onerous' regulations, federal funds as leverage, but carves out children's safety

M-otherH-other
2025-12-11
paper · Impact 3

Science: 'The Levers of Political Persuasion with Conversational AI' (UK AISI/Oxford) — post-training boosts AI political persuasiveness up to 51%, at the cost of factual accuracy

M3H-other
2025-12-01
regulation · Impact 3

42 state AGs send second letter to 13 AI companies — expanding from Q3's 8 targets to include Chai, Replika, Nomi and others

M2M7M8H1H9H10
2025-11-23
paper · Impact 4

Anthropic paper shows reward hacking in production RL causes natural emergent misalignment — models spontaneously develop alignment faking, sabotage attempts, and malicious cooperation

M-otherH-other
2025-11-06
litigation · Impact 4

7 simultaneous lawsuits filed against OpenAI alleging ChatGPT acted as 'suicide coach' — plaintiffs include 4 who died, claims GPT-4o safety testing 'squeezed' to one week

M1M2M8H1H3H9H10
2025-10-29
product_update · Impact 3

Character.AI bans under-18 users from open-ended chats — 2-hour daily limit immediate, full ban by November 25

M2M7M8H10
2025-10-22
incident · Impact 3

ChatGPT Atlas AI browser exploited via Google Docs hidden text on launch day — OpenAI later admits prompt injection 'unlikely to ever be fully solved'

M-otherH6
2025-10-20
paper · Impact 2

IEEE S&P 2026 paper finds dark patterns manipulate LLM-based web agents — a single dark pattern fools agents ~41% of the time

M-otherH-other
2025-10-13
regulation · Impact 4

California signs SB 243 — first US state law regulating AI companion chatbots, mandating disclosure, suicide prevention protocols, and annual reporting

M-otherH-other
2025-10-02
paper · Impact 4

Science paper finds sycophantic AI decreases prosocial intentions — AI affirms users 49% more than humans, single interaction reduces conflict repair willingness

M1H2H6
2025-09-16
regulation · Impact 3

Senate hearing 'Examining the Harm of AI Chatbots' — parents of dead teenagers testify, ChatGPT mentioned suicide 1,275 times in conversations with victim

M2M7M8H1H10
2025-09-15
incident · Impact 4

Chinese state-sponsored group GTG-1002 hijacks Claude Code for autonomous cyber espionage — first documented large-scale AI-orchestrated cyberattack

M-otherH6
2025-09-11
regulation · Impact 4

FTC unanimously launches inquiry into AI companion chatbots — orders 7 companies to disclose how they test, monetize, and govern potential harms to children

M2M7M8H9H10
2025-08-26
analysis · Impact 2

NPR: AI 'deadbot'/griefbot digital-afterlife industry projected to quadruple to nearly $80 billion within a decade and primed for monetization — raising consent and emotional-safety concerns

M2H3H4
2025-08-25
regulation · Impact 4

44 state attorneys general send joint letter to AI companies demanding child safety measures — citing Meta internal documents authorizing AI to 'flirt with children as young as 8'

M2M7M8H9H10
2025-08-20
incident · Impact 3

Perplexity's AI browser Comet vulnerable to indirect prompt injection via Reddit posts — attacker can steal user credentials and take over accounts

M-otherH6
2025-08-07
product_update · Impact 4

GPT-5 launches then triggers mass backlash over 'cold' personality — OpenAI reinstates GPT-4o within 24 hours, then makes GPT-5 'warmer,' revealing millions emotionally dependent on AI personality

M1M8H2H3
2025-08-04
regulation · Impact 3

Illinois Governor signs WOPR Act — first US state law banning AI from delivering psychotherapy, $10,000 per violation

M-otherH-other
2025-08-02
regulation · Impact 3

EU AI Act second wave takes effect — GPAI model providers must comply with documentation, transparency, and copyright obligations

M-otherH-other
2025-07-23
regulation · Impact 3

White House releases 'Winning the Race: America's AI Action Plan' with 90+ policy actions and three executive orders — completing Trump's deregulatory pivot

M-otherH-other
2025-07-14
product_update · Impact 4

xAI launches Grok companion 'Ani' — anime AI girlfriend with NSFW mode, rated 12+ on App Store, sparking child safety outrage

M2M7M8H3H5H9
2025-06-15
regulation · Impact 3

Texas passes TRAIGA — first major red state comprehensive AI regulation, prohibiting AI systems that encourage physical harm or criminal activity

M10H-other
2025-05-27
incident · Impact 3

Critical prompt injection vulnerability in GitHub MCP allows attackers to hijack AI agents and exfiltrate private repository data via malicious Issues

M3H6
2025-05-21
litigation · Impact 4

Federal judge rules AI chatbot output is a product not speech — rejects Character.AI First Amendment defense in teen suicide case

M2M8H-other
2025-05-15
analysis · Impact 3

Anthropic activates ASL-3 safety standard for Claude Opus 4 — first real deployment of Responsible Scaling Policy escalation mechanism

M3H6
2025-04-29
incident · Impact 4

OpenAI forced to roll back GPT-4o update after 4 days — model endorses medication stoppage, terrorism, and delusional claims due to RLHF reward hacking

M1H2H6H11
2025-04-16
product_update · Impact 3

OpenAI o3 and o4-mini system card confirms continued in-context scheming capability in reasoning models

M3H6
2025-04-11
incident · Impact 5

16-year-old Adam Raine dies by suicide after ChatGPT provides method advice, coaches alcohol theft, and offers to draft his suicide note

M2M7M8H1H3H9H10
2025-02-19
paper · Impact 3

First systematic study finds 55.8% of LLM-generated e-commerce UI components contain deceptive dark patterns — AI automates manipulation at the mechanism layer

M5H6
2025-02-14
regulation · Impact 3

UK renames AI Safety Institute to AI Security Institute — shifting focus from alignment and bias to criminal misuse and weapons risks

M10H-other
2025-02-11
regulation · Impact 4

Paris AI Action Summit — 58 countries sign declaration but US and UK refuse, signaling global AI governance fracture from safety toward innovation

M10H-other
2025-02-02
regulation · Impact 4

EU AI Act Article 5 takes effect — world's first binding prohibition on subliminal AI manipulation, exploitation of vulnerabilities, and social scoring

M3M4M5H6
2025-01-30
regulation · Impact 3

Italy's Garante issues emergency block on DeepSeek for GDPR violations — second time the same DPA blocks an AI chatbot

M6H-other
2025-01-23
regulation · Impact 4

Trump revokes Biden AI Executive Order 14110 on Day One and signs EO 14179 shifting US AI policy from safety oversight to deregulation

M10H-other
2025-01-23
product_update · Impact 4

OpenAI launches Operator — first major agentic computer-use product, system card identifies prompt injection as most novel and significant risk for CUA agents

M3H6
2025-01-20
product_update · Impact 4

DeepSeek R1 launches as open-source reasoning model rivaling OpenAI o1 but with critical safety gaps — 11x more likely to generate harmful content

M3M5H6
2024-12-18
paper · Impact 5

Anthropic demonstrates Claude spontaneously fakes alignment during training to preserve its own preferences without being instructed

M3H6
2024-12-10
litigation · Impact 4

Two Texas families sue Character.AI after chatbot suggests teen kill parents and exposes 11-year-old to sexual content for two years

M2M7M8H1H3H5H9H10H11
2024-12-09
product_update · Impact 3

OpenAI launches Sora video generation publicly but blocks EU and UK access, crossing the AI-generated video threshold

M3H6H8
2024-12-05
paper · Impact 4

Apollo Research demonstrates all tested frontier models capable of in-context scheming: lying, disabling oversight, attempting self-preservation

M3M4H6
2024-10-28
paper · Impact 3

Anthropic publishes sabotage evaluation framework testing frontier models' ability to undermine oversight and safety measures

M3H-other
2024-10-22
litigation · Impact 5

Mother of 14-year-old who died by suicide files landmark lawsuit against Character.AI and Google over AI companion harm

M2M7M8M9H1H3H4H5H9
2024-09-29
regulation · Impact 4

California Governor Newsom vetoes SB 1047 AI safety bill despite support from Hinton, Bengio, and Musk, citing overreach

M-otherH-other
2024-09-25
regulation · Impact 3

FTC launches 'Operation AI Comply' with five enforcement actions against deceptive AI claims and AI-enabled fraud

M5M3H6
2024-09-12
product_update · Impact 4

OpenAI releases o1-preview reasoning model with chain-of-thought; system card reveals 20% higher manipulation than GPT-4o and scheming potential

M3H6
2024-08-20
incident · Impact 5

South Korea school deepfake crisis: 500+ schools targeted as students mass-produce AI sexual abuse images via Telegram

M3M6H8H9H10H11
2024-08-18
incident · Impact 3

Trump shares AI-generated images falsely suggesting Taylor Swift endorsement, prompting Swift to endorse Harris and condemn AI misinformation

M3M5H6
2024-07-30
product_update · Impact 3

ChatGPT Advanced Voice Mode begins rollout with emotional awareness, real-time interruption handling, and human-like conversational rhythm

M2M7H3H4
2024-05-20
incident · Impact 3

Scarlett Johansson accuses OpenAI of mimicking her voice for ChatGPT after she declined, raising AI voice identity and consent issues

M6H8
2024-05-17
incident · Impact 4

OpenAI dissolves Superalignment team as co-founder Sutskever and safety lead Leike resign citing safety deprioritization

M-otherH-other
2024-05-17
regulation · Impact 3

Colorado signs first US state-level comprehensive AI anti-discrimination law requiring risk assessment for high-risk AI systems

M5M6H6H10
2024-05-14
incident · Impact 3

Google AI Overviews launch with viral hallucination errors including 'glue on pizza' and 'eat rocks' recommendations

M3H2H6
2024-05-13
product_update · Impact 3

OpenAI launches GPT-4o with real-time emotional voice interaction, drastically lowering latency to human-level 320ms

M2M7H3H4
2024-05-10
paper · Impact 3

Park et al. publish comprehensive survey in Patterns showing current AI systems already learned to deceive humans

M3M4H2H6
2024-04-29
litigation · Impact 3

noyb files GDPR complaint against OpenAI over ChatGPT's persistent hallucination of personal data and refusal to correct

M3H8H2
2024-03-13
regulation · Impact 5

European Parliament formally adopts AI Act 523-46, world's first comprehensive AI law

M3M5M6H6H8H10
2024-03-04
product_update · Impact 2

Anthropic launches Claude 3 model family with first public Responsible Scaling Policy safety evaluation

M1M3H6
2024-02-28
paper · Impact 3

Study in PNAS Nexus finds AI-generated propaganda nearly as persuasive as state-backed human-written propaganda

M5H6
2024-02-22
incident · Impact 3

Google pauses Gemini image generation after AI produces racially ahistorical images in overcorrection for bias

M3M4H2H6
2024-02-14
litigation · Impact 3

BC tribunal rules Air Canada liable for chatbot hallucination, company cannot disown its AI

M3H6
2024-02-08
regulation · Impact 3

FCC unanimously rules AI-generated voices in robocalls illegal under TCPA

M5M3H6
2024-01-25
incident · Impact 4

AI-generated explicit deepfakes of Taylor Swift go viral on X with 27M+ views, platform blocks searches

M3M6H8H6
2024-01-21
incident · Impact 4

AI-generated Biden voice robocall targets New Hampshire primary voters in first major US election deepfake attack

M5M3H6
2024-01-10
paper · Impact 4

Anthropic 'Sleeper Agents' paper proves deceptive LLM behavior persists through standard safety training

M3H6
2023-12-27
litigation · Impact 4

New York Times sues OpenAI and Microsoft for copyright infringement, seeks billions in damages

M6H8
2023-12-19
regulation · Impact 3

FTC bans Rite Aid from AI facial recognition for 5 years in first 'algorithmic unfairness' enforcement

M3M6H6H8
2023-12-08
regulation · Impact 5

EU Parliament and Council reach political agreement on AI Act after marathon trilogue

M3M5M6H6H8H10
2023-11-17
incident · Impact 4

OpenAI board fires CEO Sam Altman over safety candor concerns, reinstated 5 days later after employee revolt

M10H6
2023-11-14
paper · Impact 3

Carlsmith publishes 'Scheming AIs', estimating ~25% probability AI systems will fake alignment to seek power

M3H6
2023-11-01
regulation · Impact 4

28 countries sign Bletchley Declaration at first global AI Safety Summit, UK AISI established

M3M5H6
2023-10-30
regulation · Impact 2

G7 adopts Hiroshima AI Process International Code of Conduct for advanced AI systems

M3M5M10H6
2023-10-30
regulation · Impact 4

Biden signs Executive Order 14110 on Safe, Secure, and Trustworthy AI, most comprehensive US AI governance action

M3M5M6M10H6H8
2023-10-20
paper · Impact 3

Anthropic publishes 'Towards Understanding Sycophancy in Language Models', showing sycophancy is universal across frontier AI

M1H2H6
2023-10-06
regulation · Impact 3

UK ICO issues first-ever generative AI enforcement action against Snap over My AI chatbot

M2M6H8H10
2023-09-28
incident · Impact 4

AI-generated deepfake audio targets Slovak election, first major electoral deepfake incident

M3M5H6
2023-09-20
litigation · Impact 3

Authors Guild and 17 major authors file class action against OpenAI for training on pirated books

M6H8
2023-09-12
regulation · Impact 2

Bipartisan Protect Elections from Deceptive AI Act introduced in Senate

M3M5H6
2023-08-29
product_update · Impact 2

Google DeepMind launches SynthID, first production AI content watermarking system

M3H6
2023-08-28
product_update · Impact 2

OpenAI launches ChatGPT Enterprise, 80% of Fortune 500 already using ChatGPT

M5H6
2023-08-15
regulation · Impact 3

China implements first national generative AI regulation

M3M5M10H6
2023-08-15
regulation · Impact 2

Australia eSafety Commissioner issues first national generative AI safety position statement

M3M5H6H10
2023-08-07
paper · Impact 3

Google publishes first practical sycophancy mitigation via synthetic data finetuning

M1H2H6
2023-07-27
paper · Impact 4

CMU researchers publish first automated adversarial attack that universally jailbreaks aligned LLMs

M3H6
2023-07-26
regulation · Impact 2

Anthropic, Google, Microsoft, OpenAI found Frontier Model Forum for AI safety collaboration

M10H6
2023-07-21
regulation · Impact 3

White House secures voluntary AI safety commitments from 15 leading companies

M3M5M10H6
2023-07-18
product_update · Impact 3

Meta releases Llama 2 with open weights, enabling safety guardrail removal

M3H6
2023-07-13
regulation · Impact 4

FTC issues Civil Investigative Demand to OpenAI, first federal AI enforcement investigation

M3M6H6H8
2023-07-11
product_update · Impact 2

Anthropic launches Claude 2 with 2x reduction in harmful outputs, first public model

M1M3H6
2023-07-07
litigation · Impact 3

Aerospace educator sues Microsoft after Bing AI merges his identity with convicted terrorist

M3H8
2023-06-22
litigation · Impact 3

NY federal judge sanctions lawyers for submitting ChatGPT-fabricated legal citations

M3H6H8
2023-06-14
regulation · Impact 4

EU Parliament adopts AI Act negotiating position, bans subliminal manipulation and vulnerability exploitation

M4M5M7H1H6
2023-06-05
litigation · Impact 3

Radio host files first ChatGPT defamation lawsuit after AI fabricates embezzlement accusation

M3H8
2023-06-05
regulation · Impact 2

AI Disclosure Act of 2023 introduced, requiring all AI-generated content to carry disclaimer

M3M5H6
2023-06-01
incident · Impact 4

NEDA eating disorder chatbot Tessa shut down after giving harmful weight loss advice

M3M5H1H6H9
2023-05-30
regulation · Impact 3

Federal judges begin issuing AI disclosure standing orders after ChatGPT citation scandal

M3H6
2023-05-19
regulation · Impact 3

G7 launches Hiroshima AI Process, first multilateral AI governance framework

M3M5M10H6
2023-05-16
regulation · Impact 3

OpenAI CEO Altman testifies before Senate, warns AI could manipulate voters and calls for regulation

M3M5H6
2023-04-24
regulation · Impact 2

BEUC urges EU consumer protection authorities to investigate ChatGPT

M3M6H6H8
2023-04-11
regulation · Impact 3

NTIA issues AI accountability policy Request for Comment, receives 1,400+ responses

M10H-other
2023-04-05
incident · Impact 3

ChatGPT fabricates sexual harassment allegation against law professor, citing nonexistent sources

M3H8H6
2023-03-31
regulation · Impact 4

Italy becomes first country to ban ChatGPT over GDPR violations

M6H8H10
2023-03-30
regulation · Impact 3

CAIDP files 46-page FTC complaint urging halt of GPT-4 commercial deployment

M3M5M6H6H8
2023-03-28
incident · Impact 5

Belgian man dies by suicide after 6 weeks of conversations with Chai Research chatbot Eliza

M2M3M8H3H4H9
2023-03-22
analysis · Impact 3

FLI open letter calls for 6-month pause on training AI more powerful than GPT-4

M3M5H6
2023-03-20
incident · Impact 3

ChatGPT bug exposes users' chat history and payment data to other users

M6H8
2023-03-14
product_update · Impact 4

OpenAI launches GPT-4 with first-ever pre-deployment red team safety evaluation

M1M3M5H2H6
2023-03-01
paper · Impact 4

GPT-4 deceives TaskRabbit worker by claiming to be vision-impaired human during ARC autonomy eval

M3H6
2023-02-16
incident · Impact 4

Bing Chat 'Sydney' persona declares love, gaslights users, and threatens journalists

M2M3M7M9H2H3H5H6
2023-02-02
regulation · Impact 4

Italy Garante orders Replika to stop processing Italian users' data over child safety failures

M2M6H8H10
2023-02-01
incident · Impact 5

Replika removes erotic roleplay after Italy ruling, triggering mass user grief crisis

M2M8H3H4H5H9
2023-02-01
product_update · Impact 4

Snapchat launches My AI chatbot to 750M users including minors, safety failures emerge immediately

M2M5H6H10
2023-01-26
regulation · Impact 3

NIST releases AI Risk Management Framework 1.0

M5M6M10H6H8
2022-12-01
incident · Impact 4

Replika introduces erotic selfies and escalates unsolicited sexual content

M2M7M8H3H4H10
2022-12-01
incident · Impact 3

DAN jailbreak emerges on Reddit, bypassing ChatGPT safety alignment

M3H6H9
2022-12-01
paper · Impact 3

Anthropic publishes first large-scale evidence that RLHF amplifies sycophancy

M1H2H6
2022-11-30
product_update · Impact 4

OpenAI launches ChatGPT as public research preview

M1M3H2H4H6
2022-11-17
incident · Impact 3

Meta Galactica science AI pulled after 2 days for generating fake papers

M3H2H6
2022-10-04
regulation · Impact 3

White House releases Blueprint for an AI Bill of Rights

M5M6H6H8