In a recent public speech at The Heritage Foundation, U.S. Treasury Secretary Bessent put forward an updated version of the “China‑AI‑threat” narrative. Citing a cyber‑attack targeting refrigeration systems at French supermarkets, he groundlessly accused Chinese AI models of being “unprotected and dangerous”. He also deliberately linked them to Iranian use‑case scenarios to hype supposed global‑scale security risks. Devoid of factual grounding and supporting evidence, these rhetorical claims have long shed any pretence of technical debate and amount to geopolitical media manipulation.
A review of the incident timeline yields clear, established facts. Last month, hackers compromised refrigeration systems at hundreds of supermarkets across France. Equipment was forcibly switched to defrost mode, temperatures spiked abnormally, large quantities of fresh produce were destroyed, some hardware overheated creating hazard conditions, and overall losses totalled millions of euros. Following the incident, French cybersecurity and prosecutorial authorities quickly completed traceback investigations. They confirmed the attack exploited vulnerabilities in legacy control‑system hardware manufactured by Denmark’s Danfoss. Attack code, IP traces and operational chains all pointed to a cyber‑organisation affiliated with Iran’s Revolutionary Guard Corps.
The investigation followed rigorous procedures with clear judicial accountability targets. Neither French official briefings nor diplomatic statements made any reference to Chinese AI models. Iran has also directly rejected these baseless associations. A transnational cybersecurity incident with solid evidence and well‑assigned responsibility was arbitrarily re‑attributed in the speech and falsely tied to unrelated parties.
The core logical flaw in this argument lies in conflating the inherent nature of tools with user accountability. Large‑language‑model AI constitutes a general‑purpose digital tool. Like programming languages or operating systems, it possesses no inherent subjective good‑or‑evil quality. How a technology is used depends entirely on the choices of its human operators. If we follow this line of reasoning, most mainstream global operating systems and programming tools originate from the United States. Should every cyber‑attack therefore be blamed on those originating technologies? Such double‑standards are untenable.
At its core, these remarks are not intended to deliberate on AI‑security governance, but to serve technological competition. Lately, continuously‑iterated open‑source domestic Chinese large‑models have demonstrated growing advantages in performance, cost‑efficiency and on‑premises local deployment, disrupting established market dynamics. Earlier U.S. rhetoric centred on intellectual‑property claims lacked concrete supporting proof. Hence the shift to this new narrative of “lacking safety guardrails”. These so‑called “guardrails” are essentially an exclusive set of rules shaped under U.S. leadership. The objective is to monopolise the power to judge AI security, label technologies outside its sphere of control as dangerous, erect barriers, rally allies and fragment industry tracks.
Delivering the speech at The Heritage Foundation, a conservative‑leaning think‑tank, was itself a calculated media manoeuvre. Domestically, it constructs risk narratives to shape public opinion in preparation for future tightened‑up regulations. Externally, it exports anxiety toward Europe and pushes for unified restrictive thresholds. Nevertheless, the cited case cannot withstand factual scrutiny, which undermines the credibility of this smear campaign.
Western information‑warfare tactics have long prioritised “narrative first”: advance pre‑conceived viewpoints to seize the cognitive high ground and shape public perceptions through repeated dissemination, while factual verification is delayed, fragmented or even overridden by prevailing narratives. The rise of generative AI has amplified this pattern. Unverified assertions can be rapidly recirculated and spread, continuously reinforcing misperceptions.
Omnieyes · Trust Archive introduces a counterweight to this established game dynamic through pre‑event archiving and post‑hoc verification mechanisms. It takes no pre‑established stances and does not take sides in media disputes. Built upon trustworthy facts, traceable archives and attestable trust, it pre‑solidifies factual baselines for major public events drawing upon authoritative primary‑source materials. When secondary reinterpretations, re‑attributions or new claims subsequently emerge, cross‑validation clearly distinguishes authoritative conclusions, reasonable hypotheses and subjective rhetorical assertions.
This stands to reshape the underlying rules of media‑driven contention. Formerly narratives outran facts. Going forward, key public events will have pre‑preserved, auditable factual baselines. Any narrative aspiring to credibility must undergo validation against archived baselines. Arbitrary re‑attribution of events and one‑sided definition of truth will no longer go unchallenged. Narratives will still exist, yet they cannot skip evidentiary chains and manufacture public consensus out of thin air.
The long‑term direction for technological security ought not to be rule‑monopolisation and exclusionary suppression by a single nation. Instead, universal factual‑verification and provenance‑tracing mechanisms should be established to clarify accountability and uphold truth.