US hardens AI stance on China as Anthropic calls for extending lead

Washington is moving to strengthen its artificial intelligence posture towards China, with lawmakers advancing new export control bills and the Trump administration deepening cooperation with leading US AI firms as concerns grow that Beijing is rapidly closing the technological gap. A senior Anthropic executive said on Wednesday that the United States holds a six-to-nine-month lead in frontier AI models over Chinese competitors, while accusing Chinese companies of using unauthorised “distillation” techniques to replicate American advances. The comments came as the White House launched a new initiative to coordinate cybersecurity vulnerabilities with AI companies, while Congress prepared legislation aimed at further restricting China’s access to advanced chips and semiconductor manufacturing equipment. “We actually do have a pretty good handle on the lead. I think the six-to-nine-month figure is kind of an average if you look at models,” Tarun Chhabra, Anthropic’s head of national security policy, said during a discussion at the Aspen Security Forum. Chhabra characterised what he described as China’s use of model distillation as “adversarial”, describing it as “extracting the most valuable IP from the models”. He added that the United States could have been “a year to 18 months ahead” if Chinese firms had not engaged in the practice. “Virtually all of China’s leading labs now are distilling not only our models but those of our peers in the United States as well,” he said. “We are now shutting down accounts that are facilitating this distillation on the order of millions of accounts per week.” Model distillation is a widely used AI training technique in which a smaller model learns from the output of larger ones. Companies such as Anthropic and OpenAI, however, have accused leading Chinese AI firms including DeepSeek and Alibaba Group Holding of using the technique without authorisation to replicate the capabilities of their frontier models. Alibaba owns the South China Morning Post. According to Chhabra, Anthropic had new evidence that Beijing-based Zhipu AI’s latest GLM-5.2 model, which he described as “probably the most advanced Chinese model on the market now”, was distilled from both Anthropic and OpenAI models. He also accused Chinese national champion DeepSeek of “continuing its adversarial campaign for distillation”. Zhipu and DeepSeek did not immediately respond to requests for comment. Last week, China warned that Anthropic’s Claude Code poses “a serious threat” to Chinese users, while Alibaba imposed a companywide ban on employees using the coding tool, citing security concerns. The developments come as US-China AI competition intensifies, with both sides trading accusations over espionage, technology theft and the security risks posed by each other’s technologies. Amid growing concerns, the White House on Tuesday launched “Gold Eagle”, a clearing house to coordinate cybersecurity vulnerabilities with leading American AI firms. A White House statement said that multiple government departments, including the Treasury, Homeland Security and Defence, have worked closely with industry partners to enable “exploit detection” and develop faster responses to cyber vulnerabilities. “Treasury, along with our partner agencies, will continue to harness frontier AI capabilities to stay ahead of our adversaries and defend the American people from emerging threats,” Treasury Secretary Scott Bessent said. The measure follows the Trump administration’s recent move to block foreigner access to Anthropic’s most advanced models, Fable 5 and Mythos 5, which forced the company to suspend all access to the models for almost three weeks. While access was later restored, the episode raised questions about reliance on US AI providers and intensified calls among allies and partners to develop their own AI models. Former US ambassador to China Nicholas Burns, speaking during the panel discussion, criticised the Trump administration’s decision to allow exports of Nvidia’s H200 chips to Beijing, calling it one of the “grave mistakes” alongside its policies on Taiwan, South China Sea and East China Sea. “On the whole, I think we’re seeing a rather weak American policy towards China,” he said. Chhabra of Anthropic advocated for maintaining export controls on advanced AI chips and manufacturing equipment, lauding the US Congress’s role in tightening the restrictions. “I’m optimistic because of the bipartisan support in Congress for strengthening some of these controls right now,” he added. Chhabra’s remarks come as three major export control bills targeting China are set to be folded into the coming National Defense Authorisation Act (NDAA), the annual defence policy bill passed by the US Congress, marking a significant step in US efforts to restrict Chinese access to cutting-edge semiconductor chips and chipmaking equipment. The AI Overwatch Act, the Match Act and the Chip Security Act will all be included in the manager’s amendment to the 2027 NDAA in the Senate, Washington news outlet Punchbowl News reported on Tuesday, citing multiple anonymous sources. Senator Jim Banks, a Republican from Indiana, confirmed in a statement that the AI Overwatch Act he introduced would be included in the cornerstone of US defence policy legislation, “ensuring that the chips powering the next generation of artificial intelligence strengthen American innovation and national security”. Introduced in the Senate in late April, the bipartisan bill would expand congressional oversight of AI chip exports to China, with the Commerce Department required to certify to Congress that exports of advanced AI chips will not undermine US supply or facilitate remote access by unauthorised parties. The Commerce Department would also be required to develop an “American Artificial Intelligence Victory Strategy” outlining a plan to “win the AI race” against China. “We are in an AI race with the Chinese Communist Party that’s powered by the advanced chips, and the winner will determine the future of global military superiority,” said Banks. The Chip Security Act, introduced in May 2025 by Senator Tom Cotton, would further tighten US export control enforcement by requiring that advanced chip exports are equipped with location-verification mechanisms that can detect chip smuggling. Meanwhile, the Match Act would restrict Chinese access to chipmaking equipment needed to produce semiconductor chips by requiring US allies to more closely align with US export controls. The bill has received pushback from industry and allied governments including the Netherlands, home to Dutch firm ASML, which produces deep ultraviolet (DUV) and extreme ultraviolet (EUV) lithography machines, over concerns that it would hit their sales.

US hardens AI stance on China as Anthropic calls for extending lead

.css-1c6uqr6{color:inherit;font-weight:inherit;font-size:inherit;font-family:inherit;line-height:inherit;overflow-wrap:break-word;}AI company urges tighter export controls, saying Chinese rivals are rapidly closing the gap through unauthorised model distillation

AI company urges tighter export controls, saying Chinese rivals are rapidly closing the gap through unauthorised model distillation