Bessent ties AI-driven attacks to Chinese models, raising the stakes in the US-China tech rivalry
Photo: Vitaliy Haiduk / Pexels
The Treasury Secretary's claims about distillation, watermarks and a French grocery store hack point toward possible sanctions, though his attack attributions remain unverified
US Treasury Secretary Scott Bessent is blaming Chinese artificial intelligence models for AI-powered attacks. A post on X summarized his position this way: Chinese models are behind all AI-based attacks.
The comments land on top of months of escalating rhetoric from Bessent about Chinese AI. Taken together, they suggest Washington may be preparing to treat certain Chinese models less like software products and more like national security problems.
From grocery store freezers to model theft
The most striking example came in remarks Bessent delivered on November 12–13, 2026. He said a Chinese AI model was used to disable refrigeration systems in roughly 150 to 700 French grocery stores.
The span of that figure is wide. The low end and the high end differ by more than a factor of four, which says something about how fuzzy the picture still is.
Bessent attributed the operation to Iranian operatives who were supposedly using the Chinese model. In this framing, the AI is the tool and Iran is the hand holding it.
No independent verification has confirmed the connections Bessent drew between the attack and a Chinese model.
His November remarks built on earlier comments from July 21–22, 2026. At that point, Bessent said the US would scrutinize Chinese AI models for potential intellectual property theft.
The concern centers on a practice he called “industrial-scale distillation.” Technically, distillation means training a smaller or cheaper model on the outputs of a more capable one. Done at scale and without permission, it can let a rival absorb much of a frontier model’s ability without paying for the original research.
AI, tech, and the markets they move—in one daily briefing.
Daily. Free. Join 34,000+ readers across crypto, finance, and policy.
Bessent suggested Chinese models such as Moonshot AI’s Kimi K3 could carry “watermarks” that reveal they were trained on outputs from US models. If such evidence surfaced, he indicated, the consequences could include sanctions or placement on the Entity List, the US trade blacklist. Those are possibilities he raised, not actions the government has taken.
A pattern of accusations, and an offer to talk
Bessent is not making these arguments in isolation. In September 2026, US agencies accused six Chinese firms of carrying out distillation attacks on American models going back to late 2024.
Moonshot AI’s Kimi K3 was singled out in connection with large-scale distillation from Anthropic’s Fable model.
The same month brought a notably different gesture. US officials proposed setting up a hotline with China to manage AI risks, including incidents involving runaway agents and cyberattacks.
What this means for AI developers and investors
The most concrete risk sits with companies tied to Chinese AI developers. If the US moves from scrutiny to sanctions or Entity List designations, firms like Moonshot AI could find it far harder to work with American partners, suppliers and customers.
US model developers face their own set of pressures. If distillation is treated as a national security issue, firms like Anthropic have stronger incentives to lock down how their models’ outputs can be harvested at scale.
The watermark idea is worth watching closely. If US officials can credibly show that a Chinese model carries traces of American model outputs, it would give regulators a technical basis for action rather than a purely political one.
The attribution question is the other key variable. Bessent’s account of the French grocery store attack has not been independently confirmed, and the reported range of affected stores remains wide.
Watch for three things next. First, whether Treasury or other agencies take formal steps against Moonshot AI or any of the six firms accused in September. Second, whether independent researchers or French authorities weigh in on the refrigeration attack. Third, whether the proposed US-China AI hotline survives an increasingly adversarial tone.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.

