US Treasury Secretary Scott Bessent threatened Tuesday that the US could sanction Chinese AI companies over alleged IP theft — not specific models. Bloomberg first reported the threat; TechCrunch confirmed Bessent’s quotes. No enforcement is in effect. What a company-level OFAC designation would mean for model and service access is a separate, consequential question — distinct from any hypothetical model ban — and one operators should be ready to answer now.

Key takeaways:

• Bessent threatened OFAC-style sanctions on Chinese AI companies over IP theft via model distillation — no enforcement is active.
• He raised whether businesses using Chinese AI should disclose that to customers — a rhetorical question, not a formal proposal.
• An OFAC company designation could affect US access to those companies’ services; timing and scope would depend on designation terms.
• A separate Axios wholesale model-ban report has been disputed by officials; treat it as unconfirmed.

What Did Bessent Actually Say?

Speaking on Fox Business Tuesday — quotes confirmed by TechCrunch — Bessent said: “If we see, especially, that overseas models are stealing from our great companies, we have the ability to sanction them because of this theft.” He said the administration is finding US LLM “watermarks” inside Chinese-origin systems, suggesting those models were trained on proprietary American AI outputs. The watermark claim has not been backed by any published technical analysis.

Whether output-trained distillation constitutes IP theft is actively debated. Bessent’s framing is an enforcement rationale, not a legal finding. Microsoft CEO Satya Nadella has argued publicly that restricting distillation sits in tension with how AI labs themselves train — a business-level counterpoint worth tracking as any formal action develops.

On Monday, Axios reported the Trump administration was considering a wholesale ban on Chinese open-source AI models. Officials have since distanced themselves from that report, per SiliconAngle. Bessent’s statement is the most direct public escalation to date.

Why a Company Sanctions Designation Matters to Operators

OFAC sanctions target entities — specific companies, not model names. If a Chinese AI company receives an OFAC designation, US persons are generally prohibited from transacting with it, which can affect service contracts, API access, and software procurement. The precise scope and timeline depend on designation terms and any Treasury guidance issued alongside.

Two risk categories are worth tracking:

• Service availability: a provider designation would require immediate compliance review of any existing contract with that company. Transition timelines depend on license terms and Treasury guidance, not on model name alone.
• Customer disclosure: Bessent publicly asked whether businesses should disclose their use of Chinese AI to customers. Ask your legal team whether that question could translate into a disclosure obligation in your context — particularly in regulated sectors. Right now it is a policy question, not a rule.

What Should Operators Do Now?

This is a keep watching posture. Bessent said “coming days or weeks” — not immediate enforcement.

Practical steps to take before that changes:

1. Map your stack: identify every Chinese-origin model or provider service in production or evaluation — Kimi K3, DeepSeek V4, Qwen, and any workflow that routes through them.
2. Ask SaaS vendors: which base model powers each AI feature? Ask vendors explicitly whether any underlying model is sourced from a Chinese provider; review token cost and vendor implications for relevant context.
3. Consult legal and compliance: if a disclosure obligation emerges, your counsel should know your current Chinese AI exposure before it becomes urgent.
4. Document evaluations: record model names, data scope, and the rationale for any Chinese AI tests. Documented decisions are far easier to certify or migrate than undocumented use.

Watch for: a formal OFAC company designation naming specific entities, a signed executive order, or published rulemaking from the FTC or Commerce Department.

---

FAQ

Does the Bessent statement require any immediate operator action? No enforcement is active as of July 22. The practical value of acting now is reducing scramble time: a documented stack inventory and clear model names make any future compliance review or migration significantly faster. Keep Chinese AI evaluation outside production and customer-data environments until policy direction clarifies.

Is using Kimi K3 or DeepSeek legal for US businesses right now? Yes, as of July 22. No sanctions have been issued and no formal rule has been proposed. Evaluate under your existing AI governance and vendor-risk framework. Watch the next 30 to 60 days for formal agency action, an executive order, or published regulatory guidance.