The evidence shelf

Read the argument. Check the forecast.

Primary sources beside criticism; conceptual arguments separated from dated predictions; videos selected for clarity rather than heat.

  • Forecast audit

    14 April 2026

    Before he wrote AI 2027, he predicted the world in 2026. How did he do?

    Clara Collier and Daniel Kokotajlo

    A retrospective audit of Kokotajlo’s earlier forecast, including misses and partial hits.

  • Response

    2025-12

    Response to the critique of the AI 2027 timelines model

    Daniel Kokotajlo and Eli Lifland

    The authors answer the timeline critique and clarify where they agree and disagree.

  • Review

    18 September 2025

    If someone builds it, will everyone die?

    Kelsey Piper

    A safety-informed but critical examination of the book’s strongest claims.

  • Book

    16 September 2025

    If Anyone Builds It, Everyone Dies

    Eliezer Yudkowsky and Nate Soares

    A forceful argument that sufficiently capable misaligned superintelligence would cause human extinction; it is an argument, not a dated forecast.

  • Book

    2025-09

    If Anyone Builds It, Everyone Dies

    Eliezer Yudkowsky & Nate Soares

    The strongest-form case that building superintelligence with current methods leads to human extinction, and a call to stop.

  • Critique

    19 June 2025

    A deep critique of AI 2027’s timeline models

    titotal

    A detailed challenge to assumptions and mathematics beneath the scenario’s compressed timeline.

  • Scenario

    3 April 2025

    AI 2027

    Daniel Kokotajlo, Scott Alexander, Thomas Larsen, Eli Lifland and Romeo Dean

    The primary interactive scenario, including its race and slowdown branches and supporting model.

  • Essay

    3 April 2025

    Our first project: AI 2027

    Daniel Kokotajlo

    The creators introduce the project, its purpose and how they want it read.

  • Paper

    3 April 2025

    AI 2027 scenario PDF

    AI Futures Project

    The complete scenario in a portable document.

  • Essay

    2025-04

    AI 2027

    Daniel Kokotajlo, Scott Alexander, Thomas Larsen, Eli Lifland, Romeo Dean

    A month-by-month scenario of how a race to superhuman AI could unfold by 2027, with two endings.

  • Report

    2025 (updated 2026)

    International AI Safety Report

    Yoshua Bengio (chair) and ~100 experts, commissioned by 30+ governments

    The closest thing to an IPCC-style consensus review of what advanced AI can do and what the risks are.

  • Video

    2025

    AI 2027: Daniel Kokotajlo and Scott Alexander

    Dwarkesh Podcast

    A long-form walkthrough with two of the scenario’s authors.

  • Paper

    2025

    Subjective-probability forecasts of existential risk

    Ezra Karger and collaborators

    A peer-reviewed comparison of domain experts’ and superforecasters’ estimates of existential risk.

  • Podcast

    2025

    AI 2027 report

    Lawfare Daily

    Daniel Kokotajlo and Eli Lifland explain the scenario in a policy-focused interview.

  • Essay

    26 July 2024

    AI existential-risk probabilities are too unreliable to inform policy

    Arvind Narayanan and Sayash Kapoor

    A critique of treating highly uncertain extinction estimates as policy-grade measurements.

  • Forecast

    2024-06

    Situational Awareness: The Decade Ahead

    Leopold Aschenbrenner

    A compute- and geopolitics-driven case for very rapid progress and an international race.

  • Research

    11 March 2024

    Roots of Disagreement on AI Risk

    Forecasting Research Institute

    A structured investigation of why informed people remain far apart on A.I. risk.

  • Video

    2024

    Intro to AI Safety, Remastered

    Robert Miles

    A clear, non-alarmist introduction to the mechanics behind the safety argument.

  • Report

    30 May 2023

    Statement on AI Risk

    Center for AI Safety (signed by Hinton, Bengio, Altman, Amodei, Hassabis and hundreds more)

    One sentence: mitigating the risk of extinction from AI should be a global priority alongside pandemics and nuclear war.

  • Report

    22 March 2023

    Pause Giant AI Experiments: An Open Letter

    Future of Life Institute

    Calls for a six-month pause on training systems more powerful than GPT-4 while shared safety protocols are developed.

  • Book

    2023

    The Coming Wave

    Mustafa Suleyman (with Michael Bhaskar)

    A lab co-founder argues the wave of AI and synthetic biology cannot be stopped, only contained, and lays out ten steps toward containment.

  • Forecast

    6 August 2021

    What 2026 Looks Like

    Daniel Kokotajlo

    An earlier dated scenario whose near-term claims can now be compared with events.

  • Book

    2021

    Atlas of AI: Power, Politics, and the Planetary Costs of Artificial Intelligence

    Kate Crawford

    Maps AI as an extractive industry — minerals, energy, labour, data — rather than a disembodied intelligence.

  • Book

    2020

    The Alignment Problem: Machine Learning and Human Values

    Brian Christian

    A reported, historical account of how machine-learning systems come to encode values, and how researchers try to keep them aligned with ours.

  • Book

    2020

    The Precipice

    Toby Ord

    Places A.I. alongside other existential risks and explains how numerical risk estimates are used.

  • Book

    2019

    Human Compatible: AI and the Problem of Control

    Stuart Russell

    Argues that the standard model of AI (optimize a fixed objective) is the root of the control problem, and proposes machines that are uncertain about human preferences.

  • Book

    2019

    Human Compatible

    Stuart Russell

    A control-problem account that also develops a more solutions-oriented model of beneficial A.I.

  • Video

    2018

    The Orthogonality Thesis

    Robert Miles

    An accessible explanation of why intelligence alone does not guarantee human-compatible goals.

  • Book

    2014

    Superintelligence: Paths, Dangers, Strategies

    Nick Bostrom

    The foundational modern account of orthogonality, instrumental convergence and loss of control.

  • Book

    2014

    Superintelligence: Paths, Dangers, Strategies

    Nick Bostrom

    The book that framed the modern debate: why a superintelligent system might be hard to control and what strategies might help.