Anthropic urges tighter chip, data controls to secure AI lead

In a recent paper, Anthropic urged U.S. and allied governments to tighten controls on advanced chips, restrict foreign access to offshore compute, and ban model distillation attacks to preserve a 12–24 month lead in frontier AI capabilities.

The company said current export controls have helped maintain the United States’ edge but are insufficient on their own. Anthropic noted that China lacks the capacity to produce enough advanced chips domestically and cannot legally obtain sufficient supplies from abroad, which has pushed some labs toward workarounds.

Anthropic identified two main workarounds. The first is illicit compute access, which includes smuggling advanced AI chips into China and providing remote access to offshore data centers. The second is illicit model access, where copies or derivatives of frontier models are extracted through techniques known as distillation and then reused to speed domestic model development.

To close those gaps, Anthropic recommended tighter enforcement against chip smuggling, stricter controls on foreign access to overseas data centers and semiconductor manufacturing equipment, and legal clarification that distillation attacks constitute illegal conduct. The company also urged improved technical capabilities for detecting and preventing such exploits and measures to promote wider adoption of trusted U.S. AI hardware and models in global markets.

The paper cites a study estimating that, if Washington tightened controls on Chinese access to U.S. compute, the United States could hold roughly 11 times as much compute as China. The authors wrote, “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. A lead that large by 2028 would be enormously advantageous.”

Anthropic warned that powerful AI systems could be used for large-scale repression and to alter global power dynamics if the balance of capabilities shifts. The paper added that the window to establish rules and technical safeguards is narrow because capabilities and risks are advancing quickly.

The recommendations are aimed at policymakers in the United States and allied governments and at industry leaders that operate data centers and global supply chains. Anthropic framed the proposals as targeted interventions intended to reduce the ability of Chinese labs to substitute illicit access for domestic supply of advanced chips and frontier models.