US Treasury threatens sanctions on Chinese AI models over distillation theft
U.S. Treasury Secretary Scott Bessent has warned that the government may implement sanctions against Chinese AI companies if they are found to be using stolen intellectual property to train their models. The regulatory threat specifically targets model distillation techniques used to replicate frontier model capabilities, marking a potential escalation in the technological competition between U.S. and Chinese labs.
Key Takeaways
- Treasury Secretary Scott Bessent claims U.S. officials are finding American model watermarks in Chinese AI outputs.
- Moonshot AI's Kimi K3, a 2.8 trillion-parameter model, recently reached parity with Anthropic’s Claude Opus 4.8 on key benchmarks.
- Anthropic recently settled a landmark copyright lawsuit for $1.5 billion after a judge ruled it illegally stored pirated books for training.
- Bilateral U.S.-China AI talks are scheduled for September 2026, with IP theft expected to be a primary agenda item.
Why It Matters
The threat of software-level sanctions marks a significant escalation from hardware export controls, potentially blocking U.S. enterprises from using high-performance, low-cost Chinese open-source models. For the streaming and video AI sector, this could restrict access to efficient architectures like Kimi K3 that currently undercut American pricing by roughly 40%. The inclusion of model distillation as a basis for IP theft remains legally contentious, as industry leaders like Microsoft's Satya Nadella have argued that training on model outputs is a common industry practice. Watch for the Treasury Department to issue formal evidence of model watermarking or specific Entity List additions before the September bilateral talks.
Additional Context
The regulatory tension follows a surge in adoption of Chinese-led open-source technology among U.S. firms. Per CryptoBriefing (July 2026), Chinese-origin models captured 46.4% of routed token usage on OpenRouter this month, fueled by the release of Moonshot AI's Kimi K3. This model, boasting 2.8 trillion parameters, has outperformed U.S. counterparts in niche areas like the Frontend Code Arena, according to Tom's Hardware (July 2026). While the Trump administration weighs a wholesale ban on these models, previous efforts to restrict open-source AI were stalled by concerns over stifling domestic innovation. Concurrently, U.S. frontier labs face their own mounting legal pressures regarding data provenance. Per Reuters (July 2026), a federal judge granted final approval on Monday to Anthropic's record-breaking $1.5 billion settlement with a class of authors. The lawsuit, which involved the unauthorized use of nearly 500,000 copyrighted books, established a critical legal distinction: while the judge ruled that training on public data may qualify as fair use, the illegal storage of pirated source files remains a point of liability. This ruling has forced American labs to begin paying individual authors roughly $3,000 per work, a development that could increase the cost of future U.S. model development compared to less-regulated international alternatives.
Read full article at techcrunch.com
Get this in your inbox → Subscribe
Enjoy our coverage?
Add StreamingMeme as a preferred source on Google to see more of our streaming news at the top of your Search results.
Add as preferred source