Because frontier models can be used for many tasks, flaws or misuse—such as generating deepfakes or enabling cyberattacks—can have outsized effects across society.
These are the most powerful, cutting-edge AI systems—like the largest chatbots and image generators—that can do many things and could carry the biggest risks if misused.
How policymakers define 'frontier' determines which companies and products fall under new rules, shaping both safety protections and competitive dynamics.
Regulators often identify frontier models using measures such as computing power used in training, model size, or performance on benchmark tests.
Developers may be asked or required to conduct safety evaluations, share results with government bodies, and disclose certain capabilities or risks before release.
A look at how Washington, the states, and other governments are approaching oversight of artificial intelligence.
Read the guide →As Washington weighs how to govern artificial intelligence, lawmakers and industry remain split over whether binding federal rules would protect the public or hinder U.S. competitiveness.
Read the brief →