Written and maintained by CASRAI Editorial Board
Last updated
What the Center for AI Safety is
The Center for AI Safety (CAIS) is a San Francisco-based non-profit research and field-building organisation. Its stated mission is “to reduce societal-scale risks associated with AI by conducting safety research, building the field of AI safety researchers, and advocating for safety standards.” CAIS is led by Dan Hendrycks as Executive and Research Director, with Oliver Zhang as Managing Director, and organises its work across research, field-building, and a shared compute cluster.
CAIS is best known publicly for a single sentence it published in May 2023: the Statement on AI Risk. But that statement is one output of a broader organisation whose day-to-day work is technical safety research and building the pipeline of researchers who do it.
The Statement on AI Risk
In May 2023, CAIS published an open letter consisting of one sentence, intended, in its own words, to “create common knowledge of the growing number of experts and public figures who also take some of advanced AI’s most severe risks seriously.” The full, exact text is:
“Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war.”
CAIS describes the statement as signed by “hundreds of AI experts and public figures.” Signatories published on CAIS’s own site include AI researchers Geoffrey Hinton and Yoshua Bengio, Turing Award recipient Joseph Sifakis, AI lab leaders including Demis Hassabis (Google DeepMind), Sam Altman (OpenAI), and Dario Amodei (Anthropic), and UC Berkeley professor Stuart Russell. CAIS’s statement page does not itself publish a running total signatory count.
Research, field-building, and the compute cluster
Beyond the statement, CAIS’s stated work falls into a few areas:
- Safety research. CAIS runs both technical research aimed at mitigating societal-scale AI risks and more conceptual research into less formalised aspects of AI safety.
- Compute access. CAIS operates a compute cluster — a dedicated cluster of 80 A100 GPUs, plus a support team — that it makes freely available to AI safety researchers. CAIS states more than 150 researchers have used it, producing roughly 100 AI safety research papers to date.
- Field-building and training. CAIS publishes AI Safety, Ethics, and Society, a textbook and non-technical online course, and states it has used it to help train more than 1,000 people. It also runs fellowship programmes, including an AI and Society Fellowship and a Philosophy Fellowship, aimed at bringing researchers from other fields into AI safety work.
- Funding and prizes. CAIS has run funded initiatives such as SafeBench, a competition it states offered $250,000 in prizes to encourage new AI safety benchmarks.
Taken together, this is a research-and-field-building organisation rather than a standards body, an evaluator, or a coalition of companies — a distinction that matters when placing CAIS alongside the other organisations covered in this cluster.
How CAIS differs from Partnership on AI and the Frontier Model Forum
CASRAI’s cluster on frontier AI safety and governance covers several organisations that are easy to conflate because they all use the language of AI safety. CAIS occupies a distinct position from two of them:
- Partnership on AI (PAI). PAI is a multi-stakeholder membership body — academic institutions, civil-society organisations, media outlets, and industry are all full members, and it organises its work around program areas like AI Safety, Labor and Economy, and Public Policy. CAIS is not a membership organisation at all; it is a research institute that funds and conducts its own work and offers infrastructure (compute, courses, fellowships) to outside researchers.
- Frontier Model Forum (FMF). The FMF is funded and governed by the frontier AI labs themselves — its membership is limited to companies that build frontier models. CAIS’s funding and governance are independent of any single AI lab, and its statement and research output are not filtered through a coalition of developers.
Put simply: PAI is a broad multi-stakeholder coalition, the FMF is an industry-only coordinating body, and CAIS is an independent research and field-building non-profit whose best-known public act was a one-sentence statement rather than a membership standard or an industry commitment.
CASRAI’s NIKOLAI dictionary standardises the frontier-AI-safety terminology that organisations like CAIS, PAI, and the FMF each use somewhat differently — useful context for readers comparing how these groups talk about the same underlying risks.
Frequently asked questions
What is the Center for AI Safety?
A San Francisco-based non-profit whose mission is to reduce societal-scale risks from AI through safety research, field-building, and advocacy for safety standards. It is led by Dan Hendrycks as Executive and Research Director.
What did the Center for AI Safety’s Statement on AI Risk say?
In full: “Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war.” CAIS published it in May 2023 and describes it as signed by hundreds of AI experts and public figures, including AI lab CEOs and Turing Award-winning researchers.
Is the Center for AI Safety the same as CAISI?
No. The Center for AI Safety (CAIS) is an independent non-profit. CAISI is NIST’s Center for AI Standards and Innovation, a US government body. The similar abbreviations refer to unrelated organisations; see CASRAI’s separate guide on what CAISI is.
Is the Center for AI Safety an AI evaluator or standards body?
No. CAIS is a research and field-building non-profit. It does not publish third-party model evaluations in the way organisations like METR do, and it does not set enforceable industry standards.
How does the Center for AI Safety differ from the Frontier Model Forum?
The Frontier Model Forum is funded and governed by frontier AI labs themselves. CAIS’s funding and governance are independent of any single AI company, and it is not a membership coalition of developers.
