A Chinese think tank representative cornered Anthropic officials at a closed-door Singapore meeting last month and pushed for access to Mythos—the AI model so good at finding software bugs that Anthropic won't release it publicly. The company said no. When officials at the White House National Security Council heard about the exchange, they reacted with alarm, the New York Times reported on Monday.
Singapore Meeting Details
The Singapore meeting was hosted by the Carnegie Endowment for International Peace. It was a Track 2 dialogue—the unofficial kind that often paves the way for formal talks. US officials told the Times the request had almost certainly been approved by Beijing.
Why Mythos Is Rattling Banks, Hospitals, and the West Wing
Anthropic unveiled Mythos on April 7. The company says it can find previously unknown vulnerabilities in every major operating system and web browser, and write working exploits for them—without human guidance. Public release is off the table. Access has been limited to the US government and roughly 40 partner organisations.
The Wall Street Journal reported that Vice President JD Vance was alarmed enough to call CEOs including Sam Altman and Dario Amodei in April. His worry: small-town banks, hospitals and water plants couldn't survive a Mythos-style cyberattack. Treasury Secretary Scott Bessent has separately warned top banking executives. The White House is now weighing an executive order to set up a formal review process for the most powerful models. Even the Pentagon, which declared Anthropic a supply-chain risk and is moving to drop the company, is using Mythos to patch federal systems, Reuters reported Tuesday.
Mythos Has Stretched America's AI Lead, and Beijing Has Noticed
US officials previously pegged the American lead over China's best models at about six months. With Mythos and OpenAI's GPT-5.5-Cyber, that gap may now be closer to a year, the Times reported.
China's response has shifted in real time. The South China Morning Post noted that early Zhihu threads dismissed Mythos as marketing hype. Then state broadcaster CCTV, on its Yuyuan Tantian account, called the model's 'unprecedented cyberattack capabilities' a genuine threat. Shares in Chinese cybersecurity firms like Qi An Xin and 360 Security Technology rallied for several days running. One Chinese analyst, cited by the Times, framed the gap in martial terms: swords on one side, a Gatling gun on the other.
What Trump and Xi Might Actually Discuss in Beijing This Week
President Trump arrives in Beijing on Thursday. Politico, citing a White House official, reported that AI is on the summit agenda. A senior US official told reporters the two sides are exploring a 'deconfliction' channel—a hotline of sorts between AI experts in each country.
Don't expect a breakthrough. Hudson Institute fellow Bill Drexel told Politico the likeliest outcome is 'a generic goodwill statement of mutual understanding with no teeth.' Anthropic declined to comment to the Times.
