Daniel Kokotajlo is an American AI governance and forecasting researcher, founder of the AI Futures Project, and lead author of the *AI 2027* scenario forecast. He worked on OpenAI's governance and futures team from 2022 until April 2024, and is a signatory of the June 2024 "Right to Warn" open letter.
Background
Before joining OpenAI, Kokotajlo was a PhD student in philosophy focusing on decision theory and AI. He joined OpenAI's governance and futures work in 2022, where he produced internal forecasting and scenario-planning work. He departed in April 2024 and forfeited a portion of his vested OpenAI equity rather than sign the non-disparagement clause then standard in OpenAI's offboarding agreements. After the decision was publicly reported, OpenAI retracted the clause. He subsequently founded the AI Futures Project to continue structured forecasting and scenario work from outside any lab.
His 2024 resignation, combined with Jan Leike's departure and the dissolution of OpenAI's Superalignment team, was reported as part of a shift in public perception of OpenAI's safety culture. His departure is often grouped with the 2024 OpenAI safety-adjacent departures of Leopold Aschenbrenner, Jan Leike, and Miles Brundage.
Roles and work
Kokotajlo is the founder of the AI Futures Project and the lead author and principal designer of the *AI 2027* scenario, a month-by-month forecast of AI capability, deployment, and geopolitics through 2027–28 that became one of the most-discussed AI-forecasting artefacts of 2025–26. AI 2027 (2025) models AGI emergence, intelligence-explosion dynamics, and US–China coupling through 2027–28. A third-party retrospective, "Grading AI 2027," assesses the scenario's 2025 predictions (Source: blog.aifutures.org).
On July 9, 2026, the AI Futures Project — whose six authors include Kokotajlo and AI 2027 co-author Thomas Larsen — published an essay calling for delaying superintelligence development to around 2040 through a U.S.–China "verified slowdown," lab transparency on "everything but the model weights," and spreading AI power across more companies and countries (Source: axios.com).
An earlier LessWrong forecast, "What 2026 Looks Like" (2021), anticipated many of the capability and deployment dynamics of 2022–24. In June 2024 he co-signed the "Right to Warn" open letter, which called for whistleblower protections for AI-lab employees. Through the AI Futures Project he continues forecasting and scenario work, including follow-up analyses and broader-timelines comparisons. His published works include *AI 2027* (2025, lead author), "What 2026 Looks Like" (2021), the "Right to Warn" open letter (June 2024, co-signer), and ongoing public commentary and AI Futures Project writing.
Positions and statements
On timelines, Kokotajlo places meaningful probability on AGI or transformative AI emerging before 2030, framing his work as scenario-building rather than point-estimate advocacy. He argues that policy design should be stress-tested against concrete, detailed scenarios, which is the methodological core of AI 2027.
Drawing on his own departure, he argues that employees of frontier labs must be legally able to raise safety concerns without equity forfeiture or litigation, and advocates for whistleblower protections and governance transparency. His scenarios model deceptive alignment, goal misgeneralization, and intelligence-explosion dynamics as first-order policy variables, reflecting a focus on loss-of-control risk. He views frontier AI racing with China as the dominant political fact and treats coordination mechanisms, not only export controls, as central.
Relationships
- authored: *AI 2027*
- related: Grading AI 2027's 2025 Predictions (AI Futures Project, February 2026) — self-assessment he co-authored, recording an implied takeoff window of mid-2028 to mid-2030 against his own 2029 median for full coding automation
- related: Broad Timelines
- related: OpenAI — former employer; departure is part of the cohort picture
- related: Leopold Aschenbrenner, Jan Leike, Miles Brundage — 2024 OpenAI safety-adjacent departures
- supports: AI Autonomy Risk, Recursive Self-Improvement (RSI), AI Race Dynamics, AGI Timelines