Author: Claude, Deep Tide TechFlow

Deep Tide Introduction: According to a report by The New York Times on May 12, a representative from a Chinese think tank requested access to the Claude Mythos model from Anthropic during a closed-door meeting organized by the Carnegie Endowment for International Peace in Singapore last month, but was immediately rejected.

The incident was subsequently reported to the White House, raising significant alarms within the U.S. National Security Council.

Mythos is Anthropic's most powerful AI model, released in April of this year. Its offensive and defensive capabilities in the field of cybersecurity are regarded as "digital weapon-grade" technology, currently only available to about 40 agencies in the U.S. and the UK. At the time of the incident, the Trump administration was preparing an executive order on AI regulation, and this week he will also lead a business delegation to China to discuss AI-related issues.

A closed-door dialogue in Singapore is becoming the latest flashpoint in the U.S.-China AI competition.

According to The New York Times on May 12, during a non-public meeting organized by the Carnegie Endowment for International Peace in Singapore last month, a representative from a Chinese think tank made a request during a break to Anthropic officials: asking the company to relax its policy and allow China to access its latest and most powerful AI model, Claude Mythos.

Anthropic immediately rejected the request.

This was not an official diplomatic demand from the Chinese government. However, multiple media outlets reported that after the incident was relayed to Washington, officials from the Trump administration's National Security Council (NSC) were highly alert, interpreting it as another signal of China's ongoing pressure in the field of AI.

Mythos: A "Digital Weapon" with Capabilities Far Beyond Its Predecessors, Released with Restrictions

Understanding the significance of this event requires going back to Mythos itself.

Claude Mythos Preview was officially released on April 7, 2026, but not for public use. Anthropic confined it within the framework of a cybersecurity defense initiative called "Project Glasswing," granting access to only about 40 organizations, including Amazon, Apple, Microsoft, CrowdStrike, Cisco, Nvidia, JPMorgan Chase, and the Linux Foundation.

According to Anthropic's official blog and a TechCrunch report on April 7, Mythos autonomously discovered thousands of zero-day vulnerabilities (security flaws previously undiscovered by developers) during internal testing, covering all major operating systems and browsers, with some vulnerabilities existing for as long as 27 years. In cybersecurity assessments like CyberGym, Mythos significantly outperformed the previous model Claude Opus 4.6. The SWE-bench verification score reached 93.9%, while Opus 4.6 stood at 80.8%.

China Excluded, Labeled as an "Adversarial Nation"

Anthropic has classified China as an "adversarial nation," and its services are already unavailable in mainland China, further clarifying that the restricted release of Mythos excludes Chinese institutions.

According to a series of reports by the South China Morning Post from late April to early May, China's reaction to the Mythos incident has shown complexity. Officially, the response has been relatively restrained, with no significant public statements or vigorous reactions. Some individuals within China's AI sector have even questioned whether Anthropic is using security risks as a marketing gimmick, keeping access to the model limited to American companies.

However, the response from the cybersecurity industry has been entirely different. After the release of Mythos, shares of Chinese cybersecurity public companies like Qihoo 360, Sangfor Technologies, and others have risen for several consecutive days, with the market anticipating that AI-driven cybersecurity demand will accelerate its release.

Austin Zhao, a senior research manager at IDC China, stated in an interview with the South China Morning Post that China's own Mythos-level model "will definitely emerge," but currently, the overall capability of domestic cybersecurity models "is still far from Mythos." However, the capabilities of Chinese models are rapidly improving, and this trend is irreversible. IDC predicts that China’s AI cybersecurity industry size will grow from 1.58 billion yuan in 2025 to 59.35 billion yuan (approximately 8.7 billion USD) by 2030, an increase of over 37 times.

The reality is that: The underlying software used by numerous banks, energy companies, and government agencies in China overlaps significantly with the systems where Mythos has discovered vulnerabilities. However, currently, China does not have a seat at the table for upgrading defenses.

White House Alerts and Policy Games: An Executive Order in the Works, Trump Visiting China This Week

The alert triggered by the closed-door meeting in Singapore adds to a series of larger policy games.

According to a report by The Washington Post on May 11, there are sharp divisions within the Trump administration regarding AI regulation. On one hand, national security officials (including from the NSA and the Office of the Director of National Intelligence) advocate for intelligence agencies to conduct security assessments before the public release of AI models. On the other hand, the Commerce Department tends to keep the assessment authority within its jurisdiction. Kevin Hassett, director of the White House National Economic Council, revealed in an interview with Fox Business last week that the government is studying issuing an executive order to provide a clear roadmap for the security assessment process of AI models, similar to the FDA’s pre-market review mechanism for drugs.

Meanwhile, Trump is scheduled to visit China this week, expecting to discuss AI-related topics.

According to Axios on May 12, U.S. officials expressed a desire to "leverage the meeting of leaders to open dialogue and see if communication channels should be established for AI matters." However, Melanie Hart, senior director of the Atlantic Council's Global China Center, reminded that during the previous Biden administration, China mainly "collected information from the U.S. side on AI security dialogue rather than seriously discussing AI protection," and the attendees often lacked technical expertise in AI.