Chinese officials have spent the past year positioning the country as a champion of open, inexpensive artificial intelligence, yet behind the rhetoric they are uneasy about Anthropic’s flagship model, Mythos. The model, described by insiders as a "frontier" AI, has demonstrated an uncanny ability to locate and exploit software vulnerabilities at a scale humans cannot match. In one government‑technology assessment, Mythos linked hundreds of flaws together and produced functional exploit code on its first try, succeeding more than 83 percent of the time.

The performance alarmed U.S. authorities, prompting Washington in June to cut off foreign access to the model altogether. That move underscored Beijing’s anxiety: the most dangerous AI tool it fears is a closed American system it cannot legally obtain.

Anthropic, the San Francisco‑based lab behind Mythos, has taken a hard line toward China. Chief executive Dario Amodei has called for high‑end chips to be kept out of the country, accused Alibaba’s Qwen team of training on outputs from Anthropic’s Claude model, and labeled the ruling Communist Party "the most capable threat" among authoritarian regimes. The company’s stance has made it the most China‑hawkish AI lab in the United States.

Meanwhile, U.S. policymakers are also tightening the reins. Treasury Secretary Scott Bessent floated sanctions last month for developers caught stealing intellectual property, after China’s Moonshot AI released a system that claimed to rival top U.S. labs. The Commerce Ministry warned Beijing would take "all necessary measures" in response. In parallel, the federal government has barred its agencies from using Anthropic’s tools, a restriction the company is now contesting in court.

The diplomatic backdrop adds urgency. President Xi is slated to visit the United States on September 24, and Chinese officials hope to set a "win‑win" tone ahead of AI talks scheduled before the summit. Yet suspicion runs both ways. Washington worries Chinese firms may be quietly training their own models on American data, a claim that has already led to concrete actions. In June, China placed export controls on two American rare‑earth companies after the Pentagon added Alibaba and BYD to a military‑linked blacklist, illustrating a pattern of retaliation where each side hits the other’s vulnerable sectors.

Beijing is not sitting idle. Sources say officials are drafting a suite of options should Washington move first, ranging from sanctions of their own to adding U.S. AI firms to a restricted‑entity list. Because Anthropic does not operate in China, any such penalty would carry limited commercial impact but would send a clear political signal.

The emerging standoff mirrors Cold‑War‑era confrontations over missiles, only now the contested weapons are advanced AI models. As leaders prepare to sit down for talks, the battle over who controls the most powerful software‑breaking technology is poised to become a defining front in the broader U.S.–China rivalry.

This article was written with the assistance of AI.
News Factory APP - agentic news to boost your SEO & AEO.