Advances in artificial intelligence (“AI”) could bring transformative changes in society. AI has the potential for immense opportunities and benefits across a wide range of sectors, from healthcare and drug discovery to public services, and it could broadly improve productivity and living standards. However, more capable AI models also have the potential to cause extreme harm. AI could be misused for more effective disinformation, surveillance, cyberattacks, and development of chemical and biological weapons. More capable models are also likely to possess unexpected dangerous capabilities not yet observed in existing models. Laws can mitigate these risks, but in doing so must identify which models pose the greatest dangers and thus warrant regulatory attention.