Daniel Kokotajlo

Daniel Kokotajlo is the Executive Director of the AI Futures Project, a Berkeley, California 501(c)(3) nonprofit he founded in 2025 following his departure from OpenAI, where he worked in the governance division from 2022 until April 2024 (see Person entry). He is tracked here as a Voice because his public output — the April 2025 narrative scenario report AI 2027, the June 2024 "Right to Warn" insider open letter, the AI Futures Project blog reaching 26,000+ subscribers, and the 2021 scenario forecast "What 2026 Looks Like" — has done more than almost any single individual's work to install specific, dated, falsifiable AI capability timelines and an AI-safety governance frame into mainstream US public discourse and into federal policy discussion.

AI 2027: the movement-shaping scenario

AI 2027, published in April 2025 by the AI Futures Project team (Kokotajlo, Eli Lifland, Thomas Larsen, and Romeo Dean, with editing by Scott Alexander), is the most widely-read document produced by the AI-safety movement in the year of its publication. Over a million people visited the scenario website in its first few weeks — including US Vice President JD Vance and leaders of major AI companies — making it the first AI-safety document to achieve mainstream political readership at that scale. It earned Kokotajlo a place on the TIME100 AI 2025 list, following his inclusion on TIME100 AI 2024 in September 2024.

The scenario presents a modal AGI year of 2027, with the four co-authors' personal medians ranging from 2028 to 2032. Its narrative structure — a month-by-month account of how "a US AGI company" builds a superhuman coder, how China steals the model weights, and how accelerated capability research unfolds through government intervention — diverged deliberately from the abstract-risk framing common in AI-safety literature, rendering the risk legible to policymakers whose default vocabulary is geopolitics and national security rather than alignment research. The scenario's five research supplements (Compute, Timelines, Takeoff, AI Goals, and Security Forecasts) and its interactive tabletop exercise, tested across 30+ iterations with hundreds of participants including congressional staffers and researchers from OpenAI, Anthropic, and Google DeepMind, gave it the evidential scaffolding that distinguished it from mere speculation.

In November 2025, Kokotajlo publicly revised his AGI median from 2027 to the 2030s, citing updated information. The revision was documented in the AI Futures Project blog and cited as evidence that his forecasting posture is honest and falsifiable rather than advocacy-shaped — a key element of the credibility the AI 2027 voice carries in policy circles.

The Right to Warn initiative

On 4 June 2024, Kokotajlo was one of seven named signatories of "A Right to Warn about Advanced Artificial Intelligence" — co-signed with Jacob Hilton, Ramana Kumar, Neel Nanda, William Saunders, Carroll Wainwright, and Daniel Ziegler, and publicly endorsed by Geoffrey Hinton, Yoshua Bengio, and Stuart Russell (see event entry). The letter argued that frontier AI companies "have strong financial incentives to avoid oversight" and called for: the right to disclose risk information to regulators without employer reprisal; protection of anonymity for employees raising safety concerns internally; and avoidance of agreements that prevent disclosure of AI risks. The letter's four principles became the normative baseline against which subsequent corporate NDA and non-disparagement policies have been measured in public discourse.

Kokotajlo co-organised the letter as an active participant in the movement's insider whistleblowing from frontier labs strategy — grounded in his own April 2024 resignation from OpenAI and his refusal to sign the non-disparagement agreement at a personal cost of approximately $2 million in vested equity. The whistleblower identity is load-bearing in his Voice: where most AI-safety voices speak from outside the frontier labs, Kokotajlo speaks with the moral authority of having paid the exit price publicly.

Forecasting canon: from "What 2026 Looks Like" to AI 2040

Kokotajlo's earliest widely-cited public output is "What 2026 Looks Like" (2021), a scenario forecast for 2022–2026 published on the AI Alignment Forum. Asterisk Magazine's retrospective evaluation found it "prescient", noting its anticipation of chatbot proliferation and inference-time scaling years before these became common terms in public AI discussion. The held-up forecast functions as the credibility anchor for AI 2027: an author who called 2022–2026 correctly earns a hearing for 2027–2032.

In July 2026, Kokotajlo and the AI Futures Project published AI 2040: Plan A — a follow-up scenario presenting a more optimistic path for US policymakers, built around proactive governance choices that could change the AI 2027 ending. The two documents together constitute a policy-intervention frame: the catastrophic scenario and the preventative alternative addressed side by side to decision-makers rather than researchers.

The AI Futures Project Substack blog — reaching 26,000+ subscribers — is the ongoing public-output channel, carrying quarterly timelines updates, post-mortem scoring of past predictions, and model updates that document the forecasting methodology's evolution in real time.

Signature framings

Two framings in Kokotajlo's public output have travelled beyond the AI-safety research community into mainstream political and media discourse.

• "AI has a 70% chance of going horribly wrong." Kokotajlo's headline probability estimate — cited in his appearance on The Diary of a CEO with Steven Bartlett — compressed the AI 2027 scenario's catastrophic-path probability into a single striking figure. The number became the public shorthand for the scenario's key claim, carried into mass-audience podcast circuits including the Joe Rogan Experience (Episode #2551) and mainstream press coverage. It is the rhetorical move that made the AI 2027 analysis legible to audiences for whom AGI timeline arguments are opaque.

• "Everyone needs to be waking up and paying attention to this." Kokotajlo's stated urgency frame, quoted in his TIME100 AI 2025 profile by Billy Perrigo, regarding the risks of superintelligence and the inadequacy of commercial self-regulation alone. The formulation is the Voice's cleanest single statement of its theory of change — public alarm as a precondition for governance, not as a substitute for it.

Why this is a Voice entry

A Voice entry is created here, rather than additional structure on the Person entry, because Kokotajlo's public-facing output is itself the load-bearing object the corpus needs to track: AI 2027 and the AI Futures Project forecasting canon represent a specific, novel mode of movement communication — not abstract risk arguments or policy proposals but dated, falsifiable, narrative scenarios tested with policymakers and published to mass audiences — that has proven uniquely effective at moving AI-safety concerns from researcher discourse into US federal political attention. The corpus's AI-safety / pause-movement Voice slot was previously anchored in the policy-track framing of Andrea Miotti and the parliamentary-organising work of ControlAI; Kokotajlo's entry fills the complementary forecasting-and-scenario-planning Voice slot connecting the insider AI-safety world (LessWrong, AlignmentForum) with mass public discourse (Joe Rogan, mainstream press, congressional staffers). Biographical detail, employment history, and affiliations are on the linked Person entry.