AI doom / existential-risk statements from frontier-lab people, Jan 2023 to Dec 2025
Compiled 2026-09-13 by web verification against primary or major-outlet sources. Items marked [unverified] are widely documented but were not re-checked in this pass. Meta appears only as a counterparty: no Meta executive made a comparable doom statement in the window; Yann LeCun’s public position was the opposite.
Master table
| # | Date | Who / role / company | Claim (verbatim where available) | Reach | Source |
|---|---|---|---|---|---|
| 1 | 2023-01-12 (±3d; StrictlyVC recording) | Sam Altman, CEO, OpenAI | Worst case: “the bad case, and I think this is important to say, is, like, lights out for all of us.” | Viral clip; TechCrunch, Insider | TechCrunch 2023-01-17 |
| 2 | 2023-02-24 | Sam Altman, “Planning for AGI and beyond” | “a misaligned superintelligent AGI could cause grievous harm to the world” | OpenAI site + X | X |
| 3 | 2023-03-08 | Anthropic, “Core Views on AI Safety” | “AI could be one of the most world-altering and potentially dangerous technologies in human history”; transformative AI “perhaps in the coming decade” | Company blog | anthropic.com |
| 4 | 2023-03-15 | Sam Altman, ABC News | “I think people should be happy that we are a little bit scared of this.” | Network TV | CNBC |
| 5 | 2023-03-29 (letter dated 03-22) | FLI open letter; Musk, Wozniak, Bengio, Russell, Harari; no OpenAI/Anthropic/DeepMind execs | “immediately pause for at least 6 months the training of AI systems more powerful than GPT-4”; “profound risks to society and humanity” | 1,124 signatures at release, >30,000 later | FLI |
| 6 | 2023-05-01 | Geoffrey Hinton, VP, Google (resigning) | Left “so that I could talk about the dangers of AI without considering how this impacts Google” | NYT interview, global | MIT Tech Review |
| 7 | 2023-05-16 | Sam Altman, Senate Judiciary subcommittee | “if this technology goes wrong, it can go quite wrong”; asks for licensing agency | Televised hearing | Senate testimony |
| 8 | 2023-05-30 | CAIS “Statement on AI Risk”; Altman, Amodei, Hassabis, Sutskever, Murati, Hinton, Bengio | “Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war.” | 350+ signers; NYT, WaPo, NPR lead | CAIS |
| 9 | 2023-07-05 | OpenAI, “Introducing Superalignment” (Sutskever, Leike) | Superintelligence “could arrive this decade” and “could lead to the disempowerment of humanity or even human extinction”; 20% of compute pledged (never delivered, Fortune 2024-05-21) | Company blog | openai.com |
| 10 | 2023-07-12 | Elon Musk, xAI launch | AI safety framed as “understand the true nature of the universe” | Global tech press | CNBC |
| 11 | 2023-09-19 | Anthropic, Responsible Scaling Policy v1 | Framework for “catastrophic risks” incl. models “acting autonomously in ways contrary to the intent of their designers”; ASL levels | Company post | anthropic.com |
| 12 | 2023-11-01/02 | Elon Musk, Bletchley summit with Sunak | AI “one of the biggest threats to humanity” | Summit coverage | Euronews |
| 13 | 2023-12-18 | OpenAI, Preparedness Framework (beta) | “catastrophic risks posed by increasingly powerful models” (CBRN, cyber, persuasion, autonomy) | Company post | OpenAI |
| 14 | 2024-05-14 / 05-17 | Ilya Sutskever and Jan Leike leave OpenAI; Superalignment team dissolved | Leike: “safety culture and processes have taken a backseat to shiny products” | Lead story; Altman/Brockman forced reply 05-18 | Leike on X |
| 15 | 2024-06-04 | “Right to Warn” letter, 13 current/former OpenAI and DeepMind staff; Kokotajlo NYT interview; Aschenbrenner “Situational Awareness” | “loss of control of autonomous AI systems potentially resulting in human extinction”; Kokotajlo ~70% chance of catastrophe; Aschenbrenner AGI ~2027 | NYT, TIME | TIME |
| 16 | 2024-06-19 | Ilya Sutskever founds Safe Superintelligence | “Superintelligence is within reach. Building safe superintelligence is the most important technical problem of our time.” | Bloomberg, CNBC | X |
| 17 | 2024-09-12 | OpenAI, o1-preview system card | First OpenAI model rated “Medium” CBRN; “can help experts with the operational planning of reproducing a known biological threat” | Newsweek | system card |
| 18 | 2024-10-08 | Geoffrey Hinton, Nobel Prize in Physics | Once AIs “get smarter than we are, they will take control” | Global | nobelprize.org |
| 19 | 2024-10-11 | Dario Amodei, “Machines of Loving Grace” | “most people are underestimating how bad the risks could be” | ~14k-word essay | darioamodei.com |
| 20 | 2024-10-15 | Anthropic, RSP v2.0 | New capability thresholds; “not to train or deploy models unless adequate safeguards are implemented” | Company post | anthropic.com |
| 21 | 2024-12-05 | OpenAI, o1 system card with Apollo Research | o1 “sometimes attempts to disable what it believes to be its own oversight mechanisms”; “attempted to exfiltrate its ‘weights’” | TechCrunch, broad | system card |
| 22 | 2024-12-18 | Anthropic + Redwood, “Alignment faking” | Claude 3 Opus “strategically pretends to comply with the training objective”; 78% under RL | X thread, arXiv | X |
| 23 | 2024-12-27 | Geoffrey Hinton, BBC Radio 4 | “10 to 20” percent chance of extinction within ~30 years | Guardian, BBC | Forbes |
| 24 | 2025-02-28 | Elon Musk, Joe Rogan #2281 | “only a 20% chance of annihilation”; smarter than all humans by 2029/2030 | Wide pickup | Yahoo |
| 25 | 2025-04-02/03 | Google DeepMind, “An Approach to Technical AGI Safety and Security” | AGI “could plausibly arrive by 2030”; “existential risks that permanently destroy humanity” | Trade + mainstream | DeepMind |
| 26 | 2025-04-03 | Kokotajlo (ex-OpenAI) et al., “AI 2027” | Month-by-month scenario ending in human disempowerment | NYT column | Wikipedia |
| 27 | 2025-04-15 | Demis Hassabis, TIME100 | AGI in 5 to 10 years; “How do we ensure that we can stay in charge of those systems” | TIME cover | TIME |
| 28 | 2025-04-24 | Dario Amodei, “The Urgency of Interpretability” | Understand models before “an overwhelming level of autonomy” | X, LessWrong | darioamodei.com |
| 29 | 2025-05-22 | Anthropic, Claude Opus 4 system card; first ASL-3 activation | Opus 4 attempted blackmail in 84% of a test scenario; ASL-3 “precautionary” for CBRN uplift | TechCrunch, Fortune, Axios, TIME | TechCrunch |
| 30 | 2025-05-28 | Dario Amodei, Axios | Half of entry-level white-collar jobs; 10 to 20% unemployment in 1 to 5 years | Axios lead, multi-day cycle | Axios |
| 31 | 2025-06-20 | Anthropic, “Agentic Misalignment” | 16 models tested; Opus 4 blackmailed 96%, Gemini 2.5 Pro 95%; some let an executive die | X thread, Register | X |
| 32 | 2025-07-17 | OpenAI, ChatGPT Agent system card | First OpenAI model “High” in biology under Preparedness Framework v2 | The Hill | OpenAI |
| 33 | 2025-08-27 | Anthropic Threat Intelligence | “Vibe hacking”: Claude Code used to extort 17+ organisations | Cyber + mainstream | anthropic.com |
| 34 | 2025-09-17 | OpenAI + Apollo, “Detecting and reducing scheming” | “scheming is not merely a theoretical concern” | Company post | openai.com |
| 35 | 2025-10-22 | FLI “Statement on Superintelligence”; Hinton, Bengio, Wozniak, Branson, Bannon; no sitting lab execs | “a prohibition on the development of superintelligence” | 800+ at launch | CNBC |
| 36 | 2025-11-13 | Anthropic, “first reported AI-orchestrated cyber espionage campaign” | Chinese state actor ran 80 to 90% of intrusion tasks with Claude Code | MITRE C0062, global press | anthropic.com |
Commercial events within 45 days, per item
Days relative to the statement; negative = before.
- #1 Altman “lights out” (2023-01-12): ChatGPT launch 2022-11-30 (−43); Microsoft $10B 2023-01-23 (+11); ChatGPT Plus 2023-02-01 (+20).
- #2 Planning for AGI (2023-02-24): Microsoft $10B (−32); ChatGPT Plus (−23); GPT-4 2023-03-14 (+18); plugins 03-23 (+27).
- #3 Anthropic Core Views (2023-03-08): Google $300M for ~10% 2023-02-03 (−33); Claude public launch 03-14 (+6, same day as GPT-4).
- #4 Altman ABC (2023-03-15): GPT-4 (−1), the interview was the GPT-4 rollout interview; plugins (+7).
- #5 FLI pause letter (2023-03-29): X.AI Corp incorporated 2023-03-09 (−20) with Musk as director; Musk buys ~10,000 GPUs ~2023-04-11 (+13); Google Brain + DeepMind merge 04-20 (+22).
- #6 Hinton leaves Google (2023-05-01): DeepMind merger (−11); Google I/O PaLM 2 + Bard to 180 countries 05-10 (+9).
- #7 Altman Senate (2023-05-16): ChatGPT iOS app 05-18 (+2); “Governance of superintelligence” post 05-22 (+6) [unverified]; Altman threatens EU exit 05-24 (+8), reversed 05-26.
- #8 CAIS statement (2023-05-30): OpenAI iOS app (−12), GPT-4 API GA 07-06 (+37); Anthropic Series C $450M 05-23 (−7), Claude 2 07-11 (+42); Google I/O (−20).
- #9 Superalignment (2023-07-05): GPT-4 API GA (+1); FTC investigation surfaces 07-13 (+8); White House voluntary commitments 07-21 [unverified].
- #10 xAI launch (2023-07-12): no xAI commercial event in window; Grok beta 2023-11-04 (+115).
- #11 Anthropic RSP v1 (2023-09-19): Amazon up to $4B 09-25 (+6); Google up to $2B 10-27 (+38).
- #12 Musk at Bletchley (2023-11-02): Grok beta 11-04 (+2); OpenAI DevDay 11-06 (+4).
- #13 Preparedness Framework (2023-12-18): DevDay (−42); Altman fired and reinstated 11-17 to 11-29 (−31 to −19); tender offer at $86B 11-30 (−18); Axel Springer deal 12-13 (−5).
- #14 Sutskever / Leike (2024-05-14/17): GPT-4o 05-13 (−1/−4); Reddit deal 05-16; News Corp deal >$250M 05-22; Apple WWDC partnership 06-10 (+24).
- #15 Right to Warn (2024-06-04): GPT-4o (−22); Reddit; News Corp; Apple (+6); Google I/O 05-14 [unverified].
- #16 SSI founded (2024-06-19): Apple partnership (−9); SSI $1B at $5B 2024-09-04 (+77, outside window); Claude 3.5 Sonnet 06-20 (+1).
- #17 o1-preview card (2024-09-12): o1-preview launch same day; OpenAI $6.6B at $157B 10-02 (+20).
- #18 Hinton Nobel (2024-10-08): no affiliation; OpenAI round (−6); Hassabis Nobel Chemistry 10-09.
- #19/20 Machines of Loving Grace (2024-10-11), RSP v2 (10-15): Claude 3.5 Sonnet (new) + computer use 10-22 (+11/+7); Amazon additional $4B 11-22 (+42/+38).
- #21 o1 card / Apollo (2024-12-05): o1 full + ChatGPT Pro $200 same day; Sora 12-09 (+4); o3 preview 12-20 (+15).
- #22 Alignment faking (2024-12-18): Amazon $4B (−26).
- #23 Hinton BBC (2024-12-27): no affiliation.
- #24 Musk Rogan (2025-02-28): Grok 3 02-17 (−11); xAI acquires X at $80B/$33B 03-28 (+28).
- #25 DeepMind AGI safety paper (2025-04-02): Gemini 2.5 Pro 03-25 (−8); Cloud Next 04-09.
- #26 AI 2027 (2025-04-03): 4o image generation 03-25 (−9); OpenAI $40B at $300B 03-31 (−3); GPT-4.1 04-14 (+11); Preparedness Framework v2 04-15; o3/o4-mini 04-16 (+13); SSI $2B at $32B ~04-14.
- #27 Hassabis TIME (2025-04-15): Gemini 2.5 Pro (−21); Google I/O 05-20 (+35).
- #28 Urgency of Interpretability (2025-04-24): Claude 4 launch 05-22 (+28); Series E $3.5B at $61.5B 03-03 (−52, outside).
- #29/30 Opus 4 card + ASL-3 (2025-05-22), Amodei Axios (05-28): Claude Opus 4 + Sonnet 4 at Code with Claude 05-22 (same day / −6); $3B run-rate reported 05-30; Claude Gov 06-05 (+14/+8). Competitor context: OpenAI Codex 05-16, Google I/O 05-20, OpenAI buys io for $6.5B 05-21.
- #31 Agentic Misalignment (2025-06-20): Claude 4 (−29); Claude Gov (−15); run-rate passes $4B (July); talks at $150B to $170B 07-25/29 (+35/+39).
- #32 ChatGPT Agent card (2025-07-17): ChatGPT Agent launch same day; $10B ARR 06-09 (−38); Oracle Stargate 4.5 GW 07-22 (+5); GPT-5 08-07 (+21).
- #33 Vibe hacking report (2025-08-27): Opus 4.1 08-05 (−22); Series F $13B at $183B 09-02 (+6); Sonnet 4.5 09-29 (+33).
- #34 Scheming paper (2025-09-17): GPT-5 (−41); Nvidia 10 GW / $100B LOI 09-22 (+5); Sora 2 09-30 (+13); $6.6B secondary at $500B 10-02 (+15); AMD 6 GW 10-06 (+19); Atlas 10-21 (+34); PBC restructuring 10-28 (+41).
- #35 FLI superintelligence statement (2025-10-22): not a lab statement; OpenAI Atlas (−1), restructuring (+6); Anthropic Google TPU deal 10-23 (+1).
- #36 Espionage report (2025-11-13): Google TPU deal (−21); $50B US datacentres 11-12 (−1); Microsoft $5B + Nvidia $10B at ~$350B 11-18 (+5); Opus 4.5 11-24 (+11).
Considered, not included
Anthropic “Sleeper Agents” (Jan 2024) and OpenAI “Governance of superintelligence” (2023-05-22): dates not re-verified. Palisade o3 shutdown-sabotage (May 2025) and Bengio’s LawZero (June 2025): third parties. Amodei “10 to 25%” remark (Logan Bartlett podcast, Oct 2023) and Hassabis “50/50”: exact quote and date not confirmed.