США пригрозили санкциями китайским открытым ИИ-моделям за кражу интеллектуальной собственности

US Treasury Secretary Scott Bessent said on July 21, 2026, that Washington could impose sanctions on Chinese open-weight AI models over alleged intellectual property theft. According to TechCrunch, this is an expansion of the Donald Trump administration's campaign to slow China's progress in artificial intelligence.

What Scott Bessent said

The head of the US Treasury Department raised the possibility of sanctions directly against Chinese open AI models — that is, against the products themselves, not just companies or hardware shipments. The stated reason is the alleged theft of intellectual property from American developers. The secretary did not specify which models and companies could be affected — the TechCrunch report does not cite any specific names.

Scott Bessent has held the post of Treasury Secretary since January 2025. It is the US Treasury, through its OFAC office, that administers sanctions programs — from asset freezes to bans on American companies doing business with listed entities. That makes Bessent's words about models not abstract rhetoric, but a signal from an agency that has a ready-made toolkit.

• July 21, 2026 — Scott Bessent's statement, as reported by TechCrunch
• Target of possible sanctions — Chinese AI models with open access (open models)
• Justification — alleged intellectual property theft (IP theft)
• Context — the Trump administration's campaign to contain China's AI development

"The US could impose sanctions on

Chinese open AI models over alleged intellectual property theft," is how TechCrunch conveys the position of Scott Bessent, US Treasury Secretary.

Why the US is pressuring Chinese AI

Bessent's statement continues Washington's systematic policy of containing Chinese AI, TechCrunch notes. Until now, the main tools have been export controls on advanced chips — in effect since 2022 and tightened several times — and restrictions on Chinese tech companies. Sanctions against the models themselves are a qualitatively new level: the target is not hardware but a software product that is already freely distributed around the world.

Accusations of intellectual property theft have been leveled at Chinese AI labs for years. The loudest episode came in January 2025, when OpenAI claimed that DeepSeek may have used "distillation" of its models — training on the outputs of someone else's system. Now that argument is being voiced as a possible basis for government sanctions.

What DeepSeek and open weights have to do with it

The definition of open Chinese AI models primarily covers systems like DeepSeek and Qwen from Alibaba — their weights are published for free, and any developer can run the model on their own server. It is precisely this free availability that made Chinese models mainstream: after the release of DeepSeek R1 in January 2025, open models from China took hold in startups and corporate products around the world, including the US.

Exactly what sanctions against freely downloadable software might look like is not disclosed in the TechCrunch report. A ban on using open weights inside the US would have to be enforced at the level of companies, cloud providers, and app stores — there has been no precedent of this scale in the AI industry. A similar dilemma has already arisen with TikTok: banning a massively used digital product is legally difficult and politically costly — and open weights, unlike an app, are not tied to any single distributor at all.

What this means

The US is moving from chip restrictions to pressure on Chinese software itself. If Bessent's threat becomes policy, American companies will have to reconsider their use of DeepSeek, Qwen, and other open models, corporate lawyers will get a new line in their compliance checklists, and the split of the global AI market into two incompatible circuits will accelerate.