Anthropic Urges U.S. to Lock 12–24 Month AI Lead Over China

Anthropic published a policy paper urging the U.S. government to adopt three measures to preserve a 12–24 month lead in frontier artificial intelligence over China. The company called for tighter export and data‑center controls, a legal ban on model distillation attacks, and efforts to promote U.S. AI hardware and models abroad.

The paper says current export restrictions have helped maintain an American advantage but are not sufficient to guarantee a lasting gap. It identifies two workarounds that have allowed some Chinese labs to remain competitive: illicit compute access, including smuggled advanced chips and use of offshore data centers, and illicit model access, where engineers extract smaller or derivative models from larger frontier systems and reuse them to accelerate research.

To block those routes, the paper recommends three policy actions. First, strengthen enforcement against chip smuggling, limit foreign access to sensitive data centers, and restrict exports of semiconductor manufacturing equipment to reduce how much compute China can deploy. Second, clarify in law that model distillation attacks are illegal and improve technical methods to detect and prevent model exfiltration and illicit reuse. Third, promote adoption of U.S.-certified AI hardware and models among democratic partners so advanced tools are less available to the Chinese state in key commercial and strategic markets.

The paper cites a study estimating that tighter controls on access to U.S. compute could leave America with up to 11 times more compute than China. It describes how smuggled chips are integrated into local or offshore facilities and how distillation lets engineers recreate high‑performance models without direct access to the originals.

The paper states, “But if the U.S. and its allies act now to address both issues, it may be possible to lock in a 12–24 month lead in frontier capabilities.” It says the window to set rules and block illicit access is limited.

The paper warns AI could “be used to repress citizens” on an unprecedented scale and potentially reshape the global balance of power. It recommends coordinated policy across allies to raise the cost for the Chinese state to acquire frontier compute and models.