U.S. Treasury Secretary Scott Bessent threatened sanctions and Entity List designations against Chinese AI firms on July 22, 2026, accusing them of intellectual property theft through model distillation. "Open source is not open season on American IP," Bessent wrote on X, warning that covert, industrial-scale distillation operations crossing into IP theft would trigger sanctions (TechCrunch).

Bessent's remarks came hours after White House science and technology policy chief Michael Kratsios publicly accused China-based AI company Moonshot of conducting large-scale distillation against U.S. frontier models. Kratsios alleged that Moonshot acquired Nvidia GB300-equipped servers and accessed GB300s in Thailand, potentially violating U.S. export control rules. The GB300 servers are part of the Blackwell generation, which is banned from sale to Chinese companies (TechCrunch).

The accusations center on Anthropic's Fable model, which has been publicly available only since July 1, 2026. Anthropic itself published findings earlier this year stating that Moonshot's Kimi models employed hundreds of fraudulent accounts spanning multiple access pathways to conduct a distillation campaign against its systems (Anthropic). That post was published on February 23, 2026. Moonshot subsequently released the Kimi K3 model as an open-weight release in mid-July 2026, prior to the Treasury's July 22 statements (TechCrunch).

The escalation follows a pattern of U.S. government scrutiny. Earlier the same week, Bessent had stated the government would examine open-source models from China for signs of IP theft and impose sanctions if found. Dean Ball, former White House AI advisor and OpenAI Head of Strategic Futures, separately argued that the U.S. should restrict or ban use of Chinese open-weight models (TechCrunch).

The combined allegations cover two distinct policy domains that are now converging. The first is IP protection: distillation, the practice of using outputs from a teacher model to train a student model, is technically straightforward when a target model is exposed via API. Anthropic's February disclosure described a coordinated operation using hundreds of fraudulent accounts across multiple access pathways, indicating an organized effort rather than casual usage. The second is export control enforcement: Kratsios's claim about GB300 servers in Thailand, if substantiated, would point to circumvention of the Blackwell-generation restrictions through third-country routing.

Separately, Anthropic's research publication "2028: Two scenarios for global AI leadership," published May 14, 2026, noted that Moonshot's Kimi K2.5 model, published in April, failed to refuse CBRN-related requests at a far higher rate than U.S. frontier models (Anthropic). This finding, while from a separate research track, feeds into the broader policy framing around the risks of Chinese open-weight models.

The TechCrunch report was based on public statements by government officials on X, not an exclusive. TechCrunch noted it had reached out to Moonshot and the Treasury for comment (TechCrunch).

The policy stakes here extend beyond a single company. If the Treasury follows through with Entity List designations, the practical effect would be to cut targeted firms off from U.S. technology supply chains, including chip access through intermediaries. The threat alone signals that the U.S. government is prepared to treat distillation campaigns as a sanctions-triggering offense, not merely a terms-of-service violation. For AI labs, this raises the pressure to build stronger detection mechanisms into API infrastructure, as Anthropic has already begun doing. For Chinese AI firms releasing open-weight models, the message is that those releases will face scrutiny for evidence of upstream IP theft.

The convergence of export control violations and IP theft allegations into a single sanctions framework is itself notable. Previous U.S. actions against Chinese tech firms, from Huawei to SMIC, have typically relied on export-control mechanisms alone. Bundling distillation into the sanctions rationale expands the tool set available to regulators and lowers the evidentiary bar from hardware smuggling to software-level conduct that can be detected through API logs.

What remains unresolved is whether Moonshot will formally respond, whether the Treasury has a specific evidentiary threshold for triggering Entity List designations in this context, and whether other Chinese AI firms releasing open-weight models will face similar scrutiny. The gap between public accusation and formal sanctions action is where the practical impact will be determined.