The struggle between the US and China over who will lead in artificial intelligence is one of the defining competitions of our time. In this first part of our exclusive interview with a key player defining Washington’s policy on AI, US undersecretary of state for economic affairs Jacob Helberg discusses the ban on foreign access to Anthropic models and how the US can maintain its edge.
The head of the US administration’s global initiative to exclude China from artificial intelligence supply chains has defended Washington’s sudden restrictions on foreign access to Anthropic’s cutting-edge models, saying they were meant to safeguard the US and its allies.
“We do not want cyberattacks on our critical infrastructure or the critical infrastructure of our allies,” Jacob Helberg, US undersecretary of state for economic affairs, told the South China Morning Post in an exclusive interview.
Helberg, who leads Washington’s Pax Silica initiative – which also counts 22 other nations, plus the European Union, as signatories – said the administration had assessed how advanced AI models could be released safely.
“We undertook a process to make sure that we can have a sensible approach to releasing that technology in a way that is safe and secure for our critical infrastructure and our financial sector,” he said.
In early June, the Department of Commerce barred foreign nationals from using Anthropic’s Fable 5 and Mythos 5 models on unspecified national security grounds. Anthropic complied with the order, though access was restored three weeks later.
Washington is trying to balance preserving its AI lead over Beijing and addressing growing concerns over the security risks associated with the technology’s rapid development and deployment.
Under Pax Silica, members are supposed to work together to develop semiconductors and boost supply chains related to AI. While China is not explicitly a target of the programme, reducing “coercive dependencies” is listed as a key objective – widely seen as a reference to the country’s dominance in rare earths, which are critical to producing chips.
To further address security concerns, the White House on Tuesday launched Gold Eagle, a clearing house to coordinate the identification of cyber vulnerabilities with leading US AI firms.
The White House said the initiative brought together multiple government departments – including the Treasury, homeland security and defence – with industry partners to improve “exploit detection” and accelerate responses.
As Washington recalibrates its approach, Chinese President Xi Jinping is set to outline Beijing’s vision for global AI governance in a speech at the annual World Artificial Intelligence Conference in Shanghai on Friday.
Helberg said America’s private sector offered a strong path to technological leadership.
He argued that while China had pursued a state-led approach built around an “entirely in-house industrial strategy” of government-funded and government-operated projects, the US and its allies were leveraging “incredible companies that build transformative products that people really love”.
“We realise that the US government is not set up in a way to compete with the Chinese tit-for-tat on that same basis,” he added.
“We decided to lean into that strength and leverage the innovative, creative capacity and dynamism of our private sector.”
Helberg said the US continued to maintain an edge in AI, citing Chinese companies’ alleged use of “model distillation” to replicate the capabilities of top US firms.
“The US is still leading, and the evidence for that is that China is still distilling our models,” he said. “We obviously very strongly oppose it.”
The Trump administration – along with Anthropic and OpenAI – has accused Chinese companies including DeepSeek and Alibaba of using model distillation without authorisation to replicate cutting-edge models. The technique involves training smaller models with outputs generated by larger ones.
The Chinese government last week warned that Anthropic’s Claude Code posed “a serious threat” to Chinese users, while Alibaba barred employees from using the coding tool over security concerns.
Alibaba owns the South China Morning Post.
A senior Anthropic executive said on Wednesday that the United States held a six-to-nine month lead in frontier AI models and described his company’s Chinese rivals’ alleged use of model distillation as “adversarial”.
“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.
Helberg acknowledged China’s growing technological capabilities, pointing to its progress in robotics and the strength of its manufacturing ecosystem.
“I think it’s very clear that China is very technologically formidable, and they have made very impressive inroads in robotics. They obviously have a very deep, vast manufacturing ecosystem,” he said.
The competition between the nations has also prompted Washington to adopt measures aimed at blocking China’s access to the most advanced AI chips.
The Trump administration has maintained tight restrictions on exports of high-end chips to China, while making a limited exception for Nvidia’s H200 graphic processing units – though Beijing has sought to discourage domestic companies from buying them.
A senior US official told lawmakers on Tuesday that Nvidia had shipped “very few” H200 chips to the Chinese mainland and Hong Kong.
“For the American people, the bottom line is very few shipments against licences for H200s and equivalents have taken place,” said Jeffrey Kessler, the US under secretary of commerce for industry and security, at a House Foreign Affairs Committee hearing.
Congress is also pursuing additional measures, with three major China-focused export control bills set to be included in the next National Defence Authorisation Act, the annual defence policy bill.
Helberg hoped that despite the intensifying rivalry, the expected US-China dialogue on AI governance would help identify areas for “constructive engagement” between the two sides.
“Our goal is to pursue these engagements in good faith and work to restate our views and hear Chinese views,” he told the SCMP.
The two countries agreed to launch formal discussions on AI governance following the meeting between Trump and the Chinese leader in Beijing in May.
However, few details have emerged since the announcement.
The Chinese embassy in Washington said last week that China and the US “need to work together” to advance AI governance and ensure the technology better served social progress and common welfare. -- SOUTH CHINA MORNING POST
