简体中文
EnglishWhat is the Center for AI Safety (CAIS) at safe.ai?
Updated
The Center for AI Safety (CAIS, pronounced "case") is a San Francisco-based research and field-building nonprofit whose mission is to reduce societal-scale risks associated with AI. It pursues that mission through three routes: conducting safety research, building the field of AI safety researchers, and advocating for safety standards. If you are looking for an organization that treats AI risk as a societal problem rather than only a technical one, CAIS is one of the groups to know.
What CAIS says it does
CAIS describes its work as spanning research, field-building, and advocacy projects. Its stated belief is that AI has the potential to profoundly benefit the world, provided it can be developed and used safely. It also frames its mission against a gap: in contrast to the dramatic progress in AI, many basic problems in AI safety remain unsolved.
The site organizes its work under a few visible areas:
- Research — safety research aimed at societal-scale risks
- Field-building projects — programs that train and connect future AI safety researchers
- Compute cluster — shared compute infrastructure for safety work
- Advocacy — promoting safety standards
The three paths to reducing risk
1. Safety research
CAIS positions research as one leg of the mission, aimed at the basic problems in AI safety that it says are still unsolved. The site also hosts a compute cluster, which supports the idea that some of this work requires infrastructure beyond individual effort.
2. Building the field
This is the most visible part of CAIS's public programming. Examples listed on the site include:
| Program | What it is |
|---|---|
| ML Safety Infrastructure | Listed under featured field-building work |
| AI and Society Fellowship | A three-month research program on the societal impacts of advanced AI and the institutions and policies that could help societies respond well |
| Philosophy Fellowship | Listed among field-building offerings |
| AI Safety, Ethics, & Society Training | A course described as training 1000+ future AI safety leaders, offering an introduction to how current AI systems work, their societal-scale risks, and how to manage them |
| Introduction to AI Safety (with Coursera) | A course covering the drivers of modern AI development, major risks such as malicious use, competitive pressures, and loss of control, and governance strategies at different scales |
The AI and Society Fellowship page notes applications close March 24, so deadlines are program-specific and worth checking directly if you intend to apply.
3. Advocating for safety standards
The third leg is advocacy — promoting safety standards rather than only producing research. This is consistent with CAIS's public positioning on extreme risk, including its Statement on AI Extinction Risk.
Who runs it
Dan Hendrycks is the Director of the Center for AI Safety and holds a PhD in Computer Science from UC Berkeley. In his words on the site: "Preventing extreme risks from AI requires more than just technical work, so CAIS takes a multidisciplinary approach working across academic disciplines, public and private entities, and with the general public."
That quote is a useful summary of how CAIS distinguishes itself: the approach is explicitly multidisciplinary, not purely technical.
When CAIS is relevant to you
- You want to work in AI safety. The field-building programs — fellowships and courses — are the most direct entry points, and they are aimed at people who may not yet be specialists.
- You want a plain-language framing of AI risk. The Coursera course and the site's "Risks from AI" section cover malicious use, competitive pressures, and loss of control.
- You are looking for an organization's stance on extreme risk. The Statement on AI Extinction Risk is a public artifact you can read and evaluate on its own terms.
- You need compute for safety research. The compute cluster is listed as part of CAIS's work.
What the site does not tell you
The page evidence does not include pricing, membership terms, or login requirements for CAIS programs and resources, so treat cost and access as unknown until you check the specific program page. Program dates and eligibility also vary by offering — the March 24 fellowship deadline is one example, not a general rule.
For anything current — application windows, course availability, or how to get involved — the reliable move is to go to safe.ai and open the specific program page rather than relying on a summary.