On November 12, U.S. Treasury Secretary Scott Bessent delivered a startling message regarding the use of Chinese artificial intelligence (AI) models by Iranian hackers. During a speech at the Heritage Foundation’s “2026 Forum” in Maryland, Bessent highlighted a concerning incident in which these models were employed in a cyber-attack on refrigeration systems at French supermarkets. He warned that the lack of security measures in Chinese AI models poses significant risks, emphasizing that the United States must accelerate its own AI development rather than slow down.
This week, the French government released the results of its investigation into the cyber-attacks that affected around 700 retail locations throughout October. The French authorities have attributed these attacks to the Iranian Revolutionary Guard Corps, which has raised alarms about the potential threats presented by foreign cyber capabilities.
In recent months, American cybersecurity firms have observed a spike in hacking incidents involving Chinese-developed AI models such as DeepSeek and Zhihua GLM. These attacks have targeted various sectors globally, including finance, semiconductors, national defense, and government agencies, with some breaches resulting in unauthorized access to sensitive data.
Bessent made his comments in light of the ongoing concerns regarding the security vulnerabilities associated with Chinese AI models. He referenced the incident involving the French supermarket refrigeration systems, expressing disbelief that Iranian operatives were utilizing Chinese technology to execute such attacks.
“I don’t know how many of you have seen the reports about the refrigerators of 150 grocery stores in France being shut down,” he stated. “That was done using a Chinese model. This is perhaps the worst possible combination—Iranians using a Chinese model. Isn’t that a perfect reflection of our current world?”
Bessent cautioned that the deficiencies in the security protocols of Chinese AI models could lead to severe global repercussions. He predicted, “So, I foresee that you are likely to see a significant incident occur somewhere in the world.”
While Bessent did not elaborate on how Iranian hackers obtained or utilized the Chinese models, multiple organizations have issued warnings about the implications of this technology.
On the same day, a cybersecurity team named Unit 42, part of the American firm Palo Alto Networks, released a report detailing that hackers had exploited Chinese models, specifically DeepSeek and Zhihua GLM, in at least 25 attacks against 19 organizations between August and September. Ten of those organizations reported unauthorized access or data breaches.
Researchers discovered that attackers were issuing commands in Chinese for the AI models to execute cyber-attack tasks, and both models did not reject these requests.
Furthermore, another American cybersecurity firm, CrowdStrike, disclosed on October 29 that the Chinese hacking group VAPOR PANDA had used DeepSeek and AI proxy tools to target over 200 organizations worldwide. Their victims included foreign ministries, U.S. semiconductor and defense contractors, think tanks, and telecommunications service providers across various regions.
The French Ministry of Foreign Affairs issued a statement on November 10, reporting that on the night of October 11 to early hours of October 12, hackers infiltrated the control systems of refrigeration equipment used by over 700 retail sites, leading to food spoilage with estimated losses in the tens of millions of euros. The French National Cybersecurity Agency (ANSSI) noted that the incident triggered at least 15 fire alarms.
According to the investigation led by French authorities, after breaching the refrigeration control systems, the attackers altered the defrosting parameters of the equipment, raising temperatures to abnormal levels. This caused the refrigerators and freezers, originally intended for food preservation, to warm excessively, pushing businesses to dispose of the damaged goods.
The French government has linked the attack to the “Shahid Kaveh” cyber attack unit of the Iranian Revolutionary Guard Corps. French Foreign Minister Jean-Noël Barrot characterized this incident as an unprecedented and destructive cyber attack in France, and the government plans to pursue sanctions at the European level.
In addition to France, supermarkets in Belgium, Ireland, Sweden, Finland, and Estonia have also reported cyber intrusions affecting refrigeration systems. However, there is currently no evidence confirming that these incidents were initiated by Iranian hackers.
During his remarks at the forum, Bessent stated that China is striving to replicate American AI technologies. He described China’s approach as “industrial-scale model distillation,” which he labeled as a means of “stealing” American technology.
In response to the threats posed by Chinese models, Bessent argued that robust American AI models can serve as a protective barrier for businesses, personal finances, and numerous other areas.
“They (American models) can create a defense for our enterprises, individual finances, and everything else,” he stated, underscoring the necessity of not slowing down progress in this field.
He also called for the American AI industry to communicate the value of this technology more effectively to the public.
Bessent articulated that the U.S. currently leads China in the AI sector by approximately six to twelve months, emphasizing that the capabilities of AI models can undergo significant changes in a matter of months.
He concluded by asserting that China is aggressively pursuing every potential technological advantage available.