A representative of a Chinese think tank approached Anthropic officials at a meeting in Singapore last month to insist that the company change its stance and give Beijing access to its powerful new artificial intelligence model, according to people briefed on the discussions.
Anthropic refused.
The request was not an official request from the Chinese government. But the talks in Singapore were a form of exchange that is often meant to pave the way for formal, direct diplomacy.
When White House National Security Council officials learned of the information exchange at a meeting convened by Washington’s Carnegie Endowment for International Peace, they reacted with concern. Some Trump administration officials saw it as another sign that Beijing would try all possible ways to quickly acquire the most powerful model of artificial intelligence an American company has yet produced, according to people briefed on the discussions.
However subtle, the reach is a sign of the intensifying competition between China and the United States over artificial intelligence, which a growing number of national security officials and analysts they began to compare to the nuclear arms race of the Cold War. Chinese analysts see the release of the latest models from Anthropic and ChatGPT as a significant advance in American technology that could pose a threat to China.
The escalating rivalry is an important backdrop to this week’s US-China summit. President Trump is expected to arrive in Beijing on Wednesday for a meeting with Chinese leader Xi Jinping. While the race to develop the most efficient model is unlikely to be discussed, both sides could talk about access to chips that power artificial intelligence or guardrails around its use.
Previewing Sunday’s summit during a call with reporters, a senior U.S. official said artificial intelligence and cybersecurity were top of mind in Beijing, noting concerns about the “latest models of artificial intelligence.” The official said the United States and China were exploring how to create better communication over artificial intelligence and create a “deconfliction” channel in which experts from each country could address AI risks.
In April, Anthropic announced a new AI model called Mythos. The company said it was holding it back from the public release because it was skilled at finding software vulnerabilities and could cause a cybersecurity showdown. He made the model available to the US government and more than 40 organizations and companies to identify and protect against future attacks.
The technology has set off alarm bells around the world. For rivals such as China and Russia, it showed the risk of falling behind in the race to develop powerful artificial intelligence. The systems have the potential to give the nation-state the upper hand in defending against—and spearheading—cyberattacks on a wide scale and scale.
For years, US officials have estimated that the artificial intelligence models developed by the most advanced American firms are about six months ahead of the best Chinese models.
But the latest models — OpenAI’s ChatGPT 5.5 and Anthropic’s Mythos — have drastically extended the lead, potentially by nine months to a year, according to some US government and industry officials. Other US officials were more cautious, noting that China has a long history of catching up quickly.
Last year, innovations by Chinese firm DeepSeek showed the country’s ability to close the AI gap. And DeepSeek said its new model was adapted to run on chips made by Chinese tech giant Huawei, further underscoring Beijing’s push to keep up.
Still, Chinese analysts expressed concern about the potential of the new Anthropic model. One organization, IDC China, said that Mythos posed a significant risk to Chinese companiesand that the Anthropic limits on its reach created a technological gap. Another analyst, highlighting the cyber security threat, wrote that China is sharpening its swords while the other side rolled out the Gatling gun.
Chinese analysts were particularly concerned because they see the Anthropic Society as hostile to China. The start-up is currently embroiled in suits with the Pentagon, which announced it would be removed from US government networks after a dispute over how the technology would be used.
However, since its inception, Anthropic has focused its business on US national security customers. It was the first to place its artificial intelligence models in secret US networks, for example, and has long tried to keep its technology out of reach of the Chinese. In September, Anthropic expanded restrictions on its technology to include companies whose ownership structures made them vulnerable to scrutiny “from jurisdictions where our products are not permitted, such as China, regardless of where they operate.”
Both the Chinese and U.S. governments increasingly view their AI companies — including those that make the models and the cloud companies that host the computer networks on which they run — as national assets. China has blocked Meta’s $2 billion acquisition of Chinese artificial intelligence company Manus. China has also told some of its AI start-ups that they cannot accept US investment without government approval.
Anthropic and OpenAI have restricted access to their latest models to a few companies and US government agencies. Both models have the ability to discover previously unknown vulnerabilities in computer networks.
Chinese officials argued that Anthropic and OpenAI were wrong to hold tight to the models, arguing that China needed access to them to find vulnerabilities in the software, especially to protect its own critical infrastructure.
The meeting in Singapore was organized by Carnegie under the ground rules that participants would not attribute information discussed at the meetings. A number of sessions dealt with domestic regulation of artificial intelligence.
The direct request from a Chinese think tank official was made on the sidelines of the meeting, not during one of the formal sessions, according to people briefed on the events.
Matt Sheehan, a Carnegie senior fellow and organizer of the Singapore meeting, declined to comment on the talks held at the event last month, but called the broader meeting essential for both countries. “It is critical for experts in the US and China to maintain lines of communication about potential risks with AI,” he said. “That’s why we hold these dialogues.”
Anthropic officials declined to discuss the Chinese request.
Liu Pengyu, a spokesman for the Chinese embassy in Washington, said he was not aware of the exchange of information at the Carnegie meeting. But he said China is determined to balance AI development with security.
“China is ready to strengthen exchanges and cooperation in the field of artificial intelligence with all parties, promote shared benefits through joint consultation, and work together to create an open, inclusive and mutually beneficial development environment,” he said in a statement.
US officials noted that while the overture was made by a member of a Chinese think tank, it was almost certain that the Chinese government approved and directed the report. Beijing typically exercises a strong degree of control over its think tanks, especially when they are involved in unofficial diplomacy, such as this type of exchange known as Track 2 dialogue.
While China’s official government line has been that its technology sector will be able to innovate on its own and overcome restrictions imposed by the United States, Chinese companies have been more concerned about their access to advanced chips, according to U.S. industry officials and independent researchers.
U.S. officials hope that U.S. companies will continue to delay China’s access to the most advanced chips so that U.S. spy agencies can use a new program to gain access to sensitive Chinese networks, former U.S. officials say.
Industry officials are trying to convince China to change its AI strategy and not make its most powerful new models open source for anyone to use. Putting a model that has the ability to quickly infiltrate networks into the hands of hackers could unleash chaos around the world, security experts said.
But Chinese officials remain skeptical as the United States continues to look for ways to extend its lead over competition from Beijing. US companies including Anthropic, OpenAI and Google have accused Chinese firms of trying to steal their technology by essentially copying the model’s core competencies. This was reported by the Reuters agency last month, the State Department sent a diplomatic complaint to China warning it against the practice.
Congress and the Trump administration are also increasing funding for the part of the Commerce Department that proposes controls on exports of chips and other technology.
