Anthropic Rejects Chinese Bid for Mythos AI Model

The United States holds a clear lead in frontier artificial intelligence, and that advantage is now shaping diplomatic and commercial decisions at the highest levels. Anthropic’s decision to withhold its newest model, Mythos, from Chinese entities underscores how access to powerful systems has become a contested national-security issue.

Mythos, released in April, excels at identifying software vulnerabilities. Anthropic limited distribution to the U.S. government and roughly forty approved organizations to help defenders understand emerging risks before adversaries exploit them. A Chinese think-tank representative asked Anthropic officials at a Singapore meeting last month to reconsider that limit. Anthropic declined. White House officials learned of the exchange and viewed it as further evidence that Beijing seeks rapid access to the most capable American models.

President Trump arrives in Beijing this week for talks with Xi Jinping. AI and cybersecurity sit on the agenda, though the two sides are more likely to discuss chip-export rules than model-sharing. The United States has tightened controls on advanced semiconductors for four years; Nvidia’s most powerful chips remain blocked from China. Business leaders traveling with Trump, including executives from Qualcomm, Tesla, Apple, and Boeing, focus instead on agriculture, aviation, and manufacturing deals.

Chinese analysts have warned that restricted access widens a technology gap. One commentary described the situation as China sharpening swords while the other side fields a Gatling gun. U.S. officials counter that open release of models capable of large-scale network intrusion would increase global cyber risk. Both governments now treat leading AI firms as strategic assets, with Beijing blocking certain foreign acquisitions and Washington expanding export controls through the Commerce Department.

The gap between the two countries’ best models had been estimated at roughly six months; some U.S. assessments now place it at nine to twelve months. China’s DeepSeek models, adapted for domestic Huawei chips, show continued progress, yet American companies maintain tighter controls on their latest releases. The upcoming summit offers a narrow window for establishing communication channels on AI risks, but no agreement on model access appears imminent.