France says about 700 sites hit; Bessent alleges Chinese AI role

By Sophie Lindqvist

France’s Foreign Minister Jean-Noël Barrot said about 700 sites were affected by refrigeration-control cyberattacks on the night of Oct. 11–12. He attributed the operation to Iran’s Islamic Revolutionary Guard Corps.

Treasury Secretary Scott Bessent made a separate, unverified claim on Nov. 12. At a Heritage Foundation conference in Oxon Hill, Maryland, he said a Chinese artificial intelligence model disabled refrigerators in 150 French grocery stores.

He named neither the model nor any evidence. His claim has not been independently confirmed. Reporters found no French finding identifying a Chinese model.

Treasury Secretary Scott Bessent said: “Probably the worst of all worlds, the Iranians using a Chinese model.” That adds a technology claim without explaining how any model was used. The model’s origin, configuration and role remain unestablished.

France, by contrast, disclosed investigative details supporting its attribution to Iran. Bessent supplied no comparable technical trace.

What did the refrigeration attacks do?

Attackers took over refrigeration controls through web interfaces and triggered prolonged defrost cycles, according to France’s account of the October operation.

• Temperatures reached about 50 degrees Celsius in some cases.
• Attackers disabled temperature alarms.
• Stores discarded stock after the controls failed.

Barrot made that statement on Nov. 9. He called the Oct. 11–12 operation the first large-scale sabotage cyberattack on French soil. One French supermarket owner put the total cost at tens of millions of euros.

That estimate covers one business, not the whole sector. The attack count covers about 700 sites across France.

What does France’s attribution to Iran rest on?

The Paris prosecutor opened a judicial investigation into Shahid Kaveh and two deputies. Kaveh heads what France describes as an IRGC offensive cyber unit.

French authorities cited Persian-language traces and IP addresses linked to Iran. France summoned Iran’s ambassador. Barrot promised EU sanctions.

Those steps concern Iranian responsibility. They do not establish Bessent’s claim about Chinese AI. His count of 150 stores is lower than France’s figure of about 700 sites.

The two totals are not directly interchangeable. France counts affected sites, while Bessent counted grocery stores. Neither account reconciles the difference.

How strong is the Chinese-model claim?

Bessent presented the attacks as an example of inexpensive, poorly controlled Chinese AI models. He added: “They don’t have guardrails on them.” He warned of a major incident somewhere in the world.

Neither statement supplies evidence that Chinese AI caused the refrigeration failures. Bessent offered no model name, technical trace or independent confirmation. French investigative details, as reviewed by reporters, did not identify a Chinese model.

Bessent has pressed that argument against Chinese AI for months. It is easier to test than this incident. In July, he said the United States would investigate whether Chinese developers distilled American models and could sanction them. That is a policy direction, not a completed penalty.

Three distinctions matter for readers evaluating that argument. A cheap model can be dangerous without causing a particular incident. A policy warning can be credible even when its chosen example lacks proof. Here, the technology claim and the state attribution remain separate propositions, with evidence attached only to the latter.

What could the policy dispute change?

Bessent is due to travel with President Trump to the APEC summit in Shenzhen next week. He plans to discuss AI with Chinese Vice Premier He Lifeng.

On Nov. 13, Bessent said the stalled blacklisting of Nvidia customer Megaspeed was up to the Commerce Department. He placed responsibility for the delay there without announcing a decision.

For AI developers and chip suppliers, the commercial stakes are regulatory access, not refrigeration downtime. Restrictions on Megaspeed or Chinese model developers could tighten operating conditions for the affected companies. The claims offer no pricing or capacity estimates.

The next test is evidence. French investigators have pointed to Persian-language traces and Iranian IP addresses. Bessent has not produced a comparable technical link to China. His unverified allegation may shape AI policy discussions in Shenzhen.

Shenzhen’s coming talks will test whether Bessent can turn an unverified cyber claim into concrete restrictions on Chinese model developers.