Treasury Secretary Scott Bessent has put Chinese AI developers on notice: the US may impose economic sanctions on Chinese open-source AI models accused of stealing intellectual property from American companies. The warning represents one of the most direct threats yet in the Trump administration's ongoing campaign to slow China's advances in artificial intelligence.

What's Being Threatened

Bessent's comments expand the administration's existing toolkit well beyond export controls on chips and hardware. Sanctioning AI models themselves — particularly open-weight models that are freely distributed — would be a novel and legally complex step, potentially restricting American companies and researchers from downloading, using, or building on flagged models.

The targets would likely include models from Chinese labs that Washington believes were trained on proprietary American data, code, or model outputs without authorization — a practice sometimes called "model distillation stealing" or unauthorized fine-tuning on outputs from closed commercial systems.

Why This Matters Now

The timing is significant. Chinese labs have released a string of competitive open-weight models in 2025 and 2026 that have closed the gap with leading US systems faster than many analysts expected. Models like DeepSeek's R-series drew intense scrutiny earlier this year over questions about how they achieved performance parity with American counterparts at a fraction of the reported training cost.

For the Trump administration, IP theft allegations provide a legally and politically cleaner justification for restrictions than pure national security arguments, which face more rigorous judicial scrutiny.

Implications for Startups and Developers

The practical consequences for the technical community could be significant:

  • Open-source ecosystems could fracture if popular Chinese-origin models are placed on a sanctions list, forcing developers to audit their dependency chains
  • Fine-tuning pipelines that build on flagged base models could become legally precarious for US-based companies
  • Enterprise procurement teams would need to add model provenance checks to their AI vendor due diligence processes
  • Startups that have built products on top of freely available Chinese models — common given their strong performance-to-cost ratio — may face pressure to migrate

For founders, the clearest near-term action is documenting which base models underpin their products and understanding whether those models could plausibly appear on a future restricted list.

A Broader Pattern of Pressure

This threat doesn't arrive in isolation. The administration has already tightened export controls on advanced semiconductors, restricted cloud providers from serving certain Chinese customers, and pushed allies to adopt similar chip restrictions. Targeting the models themselves — the software layer — would close a loophole that hardware controls leave open: a Chinese lab can train capable models on domestic or smuggled hardware and then distribute them globally for free.

Competitor governments are watching closely. The EU has its own concerns about AI model provenance under the AI Act's transparency requirements, and a US sanctions framework could create pressure for aligned restrictions among allied nations.

What Comes Next

No sanctions have been formally filed as of this writing — Bessent's comments amount to a public warning shot. Implementing model-level sanctions would require the Treasury Department's Office of Foreign Assets Control (OFAC) to develop new designations criteria, and would almost certainly face legal challenges over First Amendment and trade grounds.

Still, the direction of travel is clear. The competition between US and Chinese AI development is increasingly being fought through trade law, not just engineering — and founders building on any part of the open-source model ecosystem should treat policy risk as a first-class technical dependency.