What AI Risks Does the Center for AI Safety Focus On?

The Center for AI Safety (CAIS) focuses on societal-scale risks from AI—risks large enough to affect whole societies rather than only individual users or companies. CAIS describes AI as a technology that can "profoundly benefit the world" while also carrying "inherent risks, including some which are potentially catastrophic." Its stated mission is to reduce those risks through safety research, field-building, and advocacy for safety standards. This article explains what that risk framing includes and why CAIS treats it as a multidisciplinary problem.

How CAIS frames AI risk

CAIS does not present AI as either purely beneficial or purely dangerous. Its position, as stated on its site, is conditional: AI's potential to benefit the world depends on whether it can be "developed and used safely."

That framing matters because it sets the scope of CAIS's work. The organization is not focused on narrow product bugs or ordinary software failures. It is focused on risks that scale with capability—the kind that become harder to contain as systems become more powerful and more widely deployed.

CAIS also notes a gap between progress and understanding: "In contrast to the dramatic progress in AI, many basic problems in AI safety have yet to be solved." The risk concern, in other words, is not only about what AI can do today, but about the mismatch between rapid capability gains and slower progress on safety.

Categories of risk CAIS highlights

Based on CAIS's own materials, its risk focus spans several categories:

• Malicious use — AI systems used deliberately to cause harm.
• Competitive pressures — dynamics where organizations or countries feel pressured to cut safety corners to keep pace.
• Loss of control — scenarios where humans can no longer reliably direct or constrain advanced systems.
• Catastrophic and extinction-level risk — the upper end of the risk spectrum, which CAIS addresses directly through its Statement on AI Extinction Risk.

CAIS's Coursera course description names malicious use, competitive pressures, and loss of control as "major risks from AI," and pairs them with governance strategies "across different scales." That pairing is notable: CAIS treats technical risk and governance risk as connected rather than separate problems.

Current systems vs. future systems

CAIS's risk page distinguishes between current AI systems and risks that emerge as systems become more advanced. This distinction is useful for readers trying to place CAIS in the broader AI safety landscape:

Why CAIS treats this as more than a technical problem

CAIS Director Dan Hendrycks states: "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."

This is the clearest statement of why CAIS's risk focus looks broader than a typical research lab's. If extreme risk can arise from competitive pressure, misuse, or governance failure, then safety research alone cannot address it. That logic explains CAIS's three-part mission:

1. Conduct safety research — technical and conceptual work on AI safety problems.
2. Build the field — training and supporting people who work on AI safety.
3. Advocate for safety standards — pushing for norms and rules that reduce risk.

The field-building side is visible in programs like the AI Safety, Ethics, & Society training, which CAIS says has reached "1000+ future AI safety leaders," and the 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.

What "societal-scale" means in practice

The phrase "societal-scale risks" is doing real work in CAIS's framing. It signals risks whose effects are not contained within a single user, company, or transaction—but instead propagate across institutions, economies, or populations.

For a concrete way to think about the difference:

• A biased hiring tool is a serious but bounded harm.
• A system that can be cheaply repurposed for large-scale disinformation, cyberattacks, or bioweapon design assistance is a societal-scale concern.
• A system that resists human oversight at a critical moment is a loss-of-control concern.

CAIS's risk categories map onto the second and third examples more than the first. That is why its materials emphasize catastrophic potential rather than only near-term product harms.

How to use this if you're evaluating CAIS

If you are trying to decide whether CAIS's risk focus matches your own interests, a few practical checks help:

• If you care about near-term, applied AI harms (bias, privacy, misinformation in specific products), CAIS's framing will feel broader and more future-oriented than you may need.
• If you care about catastrophic or existential risk, governance, and field-building, CAIS's mission and programs are directly aligned.
• If you want technical safety research, CAIS lists research as one of its three pillars, alongside field-building and advocacy.
• If you want to get involved, the site points to a fellowship, a training course, a Coursera course, a blog, a newsletter, publications, and impact reports.

The site also provides a Statement on AI Extinction Risk and an FAQ, which are the most direct places to check CAIS's own position on the most severe risk claims rather than relying on secondhand summaries.

The short version

CAIS focuses on AI risks that are large enough to affect society as a whole—malicious use, competitive pressure, loss of control, and catastrophic outcomes—and treats them as problems that require research, a trained field of researchers, and safety standards, not technical fixes alone. Its own framing is conditional rather than alarmist: AI can benefit the world, but only if it is developed and used safely.