Center for AI Safety

We’ve released a statement on the risk of extinction from AI. Signatories include: - Three Turing Award winners - Authors of the standard textbooks on AI/DL/RL - CEOs and Execs from OpenAI, Microsoft, Google, Google DeepMind, Anthropic - Many more safe.ai/statement-on-a…

The full edition is available here, along with links to open roles at CAIS: newsletter.safe.ai/p/aisn-71-cybe…

Hackers, including a North Korea-linked group, attacked the AI supply chain, stealing data potentially worth billions. This week's AI Safety Newsletter covers the attacks, a federal judge calling the Pentagon's action against Anthropic "Orwellian", Sanders and AOC's bill to halt all new datacenter construction and chip exports, and more.

The AI Doc is now playing in US theaters, featuring CAIS’s @hendrycks alongside @Liv_Boeree and many more. AI’s biggest questions deserve public discussion. Watch the film in theaters this week, then bring the conversation to your community.

#Controlled_opposition" target="_blank" rel="nofollow noopener">en.wikipedia.org/wiki/Oppositio…

To clarify, the Center for AI Safety has not taken funding from Coefficient Giving / Open Philanthropy for years. We believe the effective altruism movement is, unfortunately, controlled opposition. The less influence it has on AI safety, the better.

@boazbaraktcs @mia_glaese @justinlinw @danboneh @davisbrownr @MantasMazeika96 @MrinankSharma @MilesMcCain newsletter.mlsafety.org/p/mlsn-19-hone…

The fellowship includes regular guest speaker events by professors at Stanford, Penn, Johns Hopkins, and more. Further details and a link to the application can be here: safe.ai/fellowship

Applications are now open for the AI and Society Fellowship at the Center for AI Safety: a fully funded, 3-month summer fellowship in San Francisco for scholars in economics, law, IR, and adjacent fields to research the societal impacts of advanced AI. Apply by March 24.

@ProfArbel Thank you for sharing. The fellowship also welcomes scholars in economics, IR, and adjacent fields to study the societal implications of advanced AI. It will also host speakers from Stanford, Penn, JHU, and more. Apply by March 24. For more information: safe.ai/fellowship

Hi law profs/fellows: Interested in the big social implications of AI? Want a funded summer fellowship in SF by some of the best in the business? @CAIS holds a call for summer fellowships that allow you to pursue your research tracks with close feedback loops

More detailed results at: dashboard.safe.ai

While Gemini 3.1 Pro performs better than 3.0 Pro on a variety of safety benchmarks, its risk index remains high due to its susceptibility to jailbreaks.

New Gemini 3.1 Pro results: Large improvements in text reasoning, and the Gemini 3 series continues to dominate vision benchmarks. However, the model still has a higher risk index (worse safety) than each of its frontier peers.

Opus 4.6 is the most capable model on our text benchmarks, but less safe than its predecessor Text: #1 overall, strongest on coding and abstract reasoning Vision: #4, well behind Gemini 3 Safety: Worse than Opus 4.5 on bioweapons assistance, deception, and overconfidence

For more details, see remotelabor.ai

AI agents are getting good at coding, but how close are they to automating all digital labor? New Remote Labor Index results: Opus 4.5 is able to automate 3.75% of remote labor projects, with GPT-5.2 in second place.

Full article: nature.com/articles/s4158…

Last week, Humanity’s Last Exam was published in @Nature. In just over a year, model scores on HLE have risen from under 5% to nearly 40%. Thank you to @scale_AI and the 1000+ HLE co-authors for helping policymakers and the public track these rapid advances in AI capabilities.