US Accuses Chinese AI Firms of Model Distillation

Model distillation is a standard and often legitimate AI development tool, but the US says it is being abused at scale to copy capabilities from American leaders without permission. That raises the stakes for frontier developers that have spent billions building proprietary models, and it increases pressure on them to tighten access controls, monitor abuse and defend intellectual property.

The allegation lands just before planned US-China talks on AI safety in September and adds another flashpoint to already strained technology relations. Reuters has previously reported OpenAI accused DeepSeek of unauthorized data use, while Anthropic has warned that illegally distilled models could pose national security risks because they may not include proper safeguards.

Investors are likely to view the dispute as supportive for US AI incumbents in the near term, even as it underscores the fragility of the global AI supply chain and the risk of tighter policy action. Semiconductor and cloud names tied to the AI build-out, including Nvidia and Microsoft, remain exposed to any escalation in restrictions, while Chinese AI firms face a higher chance of scrutiny, slower commercialization and more expensive model development.

The immediate focus now shifts to whether Washington uses the allegations to justify fresh limits on Chinese access to US AI systems or to press for new verification rules in bilateral talks. Any retaliation or broader export-control response could ripple through AI software, chips and cloud infrastructure on both sides of the Pacific.