Dangerous Misuse by Bad Actors
Powerful AI can help non-experts plan cyberattacks, design fraud, manipulate markets, evade detection, and pursue biological or chemical threats.
The more capable and widely available these systems become, the more dangerous misuse becomes. Capabilities that once required a team of experts can be handed to anyone with a laptop and bad intentions.
This is why safety must be built before deployment, not after. Once a dangerous capability is released into the world, it cannot be recalled.
- AI lowers the expertise needed to cause serious harm, from cyberattacks to fraud to biological and chemical threats.
- Capabilities released openly cannot be taken back.
- The pace of release is set today by competitive pressure, not by public safety.
- Criminals and hostile actors are early adopters of every powerful new tool.
What we are asking for
The safeguards most relevant to this danger. See all six on Our Solutions.
Limits on Dangerous Capabilities
Limits on the release of dangerous capabilities, so that the most hazardous systems are held back until they can be shown to be safe.
Safety Testing Before Deployment
Independent safety testing before powerful systems are released, with independent monitoring afterward. Safety must be built before deployment, not after disaster.
Transparency from AI Developers
Transparency from AI developers about what they are building, what it can do, who it affects, and what safeguards are in place, backed by independent monitoring.
National Investment and International Cooperation
National investment in AI safety, and international cooperation to prevent a reckless race toward systems humanity cannot control.
Common questions
Isn’t this just a cybersecurity problem for companies to solve?
Would limits on dangerous capabilities stop legitimate research?
What can I do?
Explore the other challenges
Protect humanity's future.
Become a monthly supporter. Your gift funds plain-language education and public advocacy for serious safeguards.
Donations are not tax-deductible at this time.
