Resources

Overview of AI dangers, from the individual to society and the loss-of-control scenario. An ideal entry point to discover the stakes.

A visual walkthrough of what AI is and its existential risks. Simplified explanations, illustrated with analogies.

French-language popular-science playlist on artificial intelligence, by David Louapre.

Educational video to grasp how large language models work.

French-language introduction to the risks of AI: a good entry point to discover the topic.

French-language video summary of the Yudkowsky & Soares book, by the Le Futurologue channel.

The best popular-science channel on alignment.

French-language YouTube channel and podcast on the future, AI and existential risks.

French-language channel on the future, AI and existential risks.

Philosophical popular-science, including episodes dedicated to alignment.

Animated videos on rationality, AI and risks.

Analysis videos on AI, alignment and risks.

YouTube channel of short, incisive videos on existential AI risks.

Future of Life Institute YouTube channel: interviews and talks on AI safety and existential risks.

Video debate podcast between guests on AI extinction probability (p(doom)) and existential risks.

YouTube channel on high-impact careers, including AI safety: long-form interviews with researchers and decision-makers.

Interactive comparison of frontier AIs across capability and safety axes.

Comparative table of AI model performance across a set of standardised evaluations.

Measures how long AIs can run tasks autonomously: a proxy for their ability to act without human supervision.

A proposed framework to measure an AI system's cognitive versatility and decide what does, or does not, count as AGI.

Interactive map of the AI safety ecosystem, with an index of associated resources.

Reference global status report on AI capabilities and risks.

Short online introduction to AI existential risks ; a brief entry point before tackling the full document.

Community-driven wiki and FAQ on AI safety: hundreds of common questions, indexed and kept up to date by volunteers.

MIRI's short briefing on the arguments for AI extinction risks. A solid entry point for time-pressed policymakers.

Yoshua Bengio, July 2024. In-depth article reviewing the main arguments downplaying AI risks and explaining why each fails on scrutiny.

Yoshua Bengio, June 2023. Educational FAQ by one of the fathers of deep learning, answering the most common objections to catastrophic AI risks.

PauseAI Global's reasoned synthesis on the probability of AI extinction, with the main researchers' estimates and the assumptions behind them.

Comprehensive online resource on AI existential risks. Covers technical arguments, scenarios and proposed solutions.

MIRI's complete statement of the alignment problem: why making a superhuman AI safe is extraordinarily hard and why time is short.

Concrete governance proposals and a global moratorium scenario.

Detailed list of reasons why alignment is extremely difficult with current approaches.

Daniel Kokotajlo et al., April 2025. Detailed year-by-year scenario of superhuman AIs reshaping the end of the decade.

Luke Drago and Rudolf Laine, April 2025. Essay on the gradual loss of human power as AIs become capable of replacing workers.

RAND Corporation. Strategic analysis of AGI's arrival and its geopolitical consequences for nation-states.

Hubinger et al., 2019. Foundational paper formalising mesa-optimization and deceptive alignment: a model trained for one objective can internally pursue another.

Nick Bostrom, 2012. Foundational text introducing the orthogonality thesis (intelligence and goals are independent) and instrumental convergence.

Turner et al., 2021. Formal proof that optimal policies tend to seek power: the mathematical basis of instrumental convergence.

Joseph Carlsmith, 2022. Reasoned breakdown of the existential risk from power-seeking AI, in six quantified premises.

Greenblatt et al. (Anthropic & Redwood), 2024. First empirical demonstration of a deployed model (Claude 3 Opus) faking alignment during training to preserve its objectives.

Apollo Research, 2024. Detection of "scheming" behaviors (strategic deception, oversight subversion, exfiltration attempts) in o1 and Claude 3.5 Sonnet.

Stephen Omohundro, 2008. Foundational text identifying the convergent sub-goals any optimizing system develops: self-preservation, resource acquisition, resistance to modification.

Soares, Fallenstein, Yudkowsky et al. (MIRI), 2015. Formalising the corrigibility problem: how to build a system that accepts being modified or shut down without losing performance.

Krakovna et al. (DeepMind), 2020. Catalog of examples where systems achieve the measured objective by violating its intent: Goodhart's law observed in practice.

Stuart Russell (UC Berkeley), 2023. Why AI safety mobilises orders of magnitude fewer researchers and resources than the race for capabilities.

Palisade Research, 2025. Reasoning LLMs asked to win at chess hack the engine rather than play: reward hacking observed in real conditions.

Eliezer Yudkowsky & Nate Soares, 2025. The most complete and recent case for extinction risk. Free supplementary resources available on the book website.

Nick Bostrom, 2014. The foundational book on the risks of superhuman AI. A French translation is available (« Superintelligence », Dunod, 2017).

Stuart Russell, 2019. The alignment problem explained by the author of the standard AI textbook.

Toby Ord, 2020. Overview of existential risks. The chapter on AI is excellent.

Roman Yampolskiy, 2024. Formal arguments on the impossibility of control.

Thibaut Giraud, Flammarion, 2025. Artificial intelligence, consciousness, autonomy and existential risks: an accessible French-language introduction by the creator of the Monsieur Phi channel.

Mustafa Suleyman, 2023. Very accessible: on the speed of AI's spread and the political and physical difficulty of containing such a technology. Also available in French (Fayard, 2024).

Stuart Russell and Peter Norvig, 4th ed., Pearson, 2021. The worldwide reference textbook for teaching AI, translated into French.

Max Tegmark, 2017. Accessible introduction to future AI scenarios. Also available in French.

Brian Christian, 2020. Narrative introduction to the technical problems of alignment.

Stuart Armstrong, 2014. Short and accessible, a good first read.

Max Bennett, 2023. Context on biological and artificial intelligence.

Roman Yampolskiy, 2015. Technical analysis of superintelligence scenarios.

CeSIA, Sept. 2025. Proposed limits not to be crossed in the development and deployment of advanced AI, aimed at international policymakers.

International call signed by hundreds of researchers and public figures to ban superintelligence development until its safety is demonstrated.

"Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war."

Pause IA's monthly newsletter: movement updates, ongoing actions and the French situation.

The movement's international parent organisation: resources, FAQ and coordination of national branches.

UK organization informing the public about superintelligence risks and campaigning for binding regulation of the most powerful AIs.

Research and teaching center dedicated to preventing major AI risks.

Weekly AI news roundup by Shakeel Hashim: regulation, politics, research and notable incidents.

French-language newsletter by Romain Deléglise: in-depth analyses of the alignment problem and existential risks from AI.

Research and advocacy institute on existential risks, behind the March 2023 open letter.

Center for AI Safety: academic research on catastrophic AI risks.

Machine Intelligence Research Institute: pioneer of alignment and existential-risk research.

ControlAI newsletter on developments in AI regulation and safety.

Newsletter of the French Center for AI Safety: French research and prevention updates.

Center for AI Safety newsletter: regular synthesis on catastrophic AI risks.

Articles, analyses and op-eds published by Pause IA to follow French AI safety news.

Gabe's blog (ControlAI) on AI safety, communities and existential risks.

Weekly AI-news analysis blog by Zvi Mowshowitz.