Center for AI Safety (CAIS)
The Center for AI Safety (CAIS) is a nonprofit organization focused on reducing societal-scale risks from artificial intelligence. Founded by Dan Hendrycks, CAIS conducts research, provides resources for the AI safety community, and engages in public advocacy about AI risks.
History
CAIS was founded in 2022 to address what its founders saw as a critical gap in organized efforts to mitigate catastrophic AI risks. The organization gained significant public attention in May 2023 when it released the "Statement on AI Risk," a one-sentence declaration comparing AI extinction risk to pandemics and nuclear war, signed by hundreds of AI researchers and executives.
Mission and Approach
CAIS takes a multi-pronged approach to AI safety:
- Research: Conducting and funding technical AI safety research
- Field Building: Providing compute grants and resources to safety researchers
- Public Advocacy: Raising awareness about AI risks among policymakers and the public
- Benchmarks: Developing evaluations for measuring AI safety and capabilities
Key Initiatives
Statement on AI Risk (2023)
CAIS organized a widely-publicized statement declaring: "Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war." Signatories included Yoshua Bengio, Dario Amodei, Sam Altman, Demis Hassabis, and hundreds of other AI researchers and executives.
Compute Cluster
CAIS operates a compute cluster to provide researchers with resources for AI safety experiments, helping address the resource gap between safety researchers and well-funded AI labs.
ML Safety Course
CAIS developed an introductory machine learning safety course covering topics including reward hacking,goal misgeneralization, and AI governance.
Research Areas
- Dangerous capability evaluations
- Deceptive alignment detection
- Robustness and distributional shift
- AI governance and policy
- Anomaly detection in AI systems
Leadership
- Dan Hendrycks - Director