Anthropic grants Japan's three mega-banks access to Claude Mythos — first East Asian expansion

Anthropic grants Japan's three mega-banks access to Claude Mythos—first East Asian expansion

Anthropic is granting Japan's three largest banks—MUFG, Sumitomo Mitsui, and Mizuho—access to Claude Mythos, marking the first time an East Asian company has been cleared to use the model. The banks were informed of the decision by Treasury Secretary Scott Bessent during a meeting in Japan on Tuesday, ahead of the Trump-Xi summit in Beijing.

Mythos had been restricted to roughly 50 organizations worldwide, including U.S. firms, U.S. banks, and U.K. government entities. The move signals a deliberate expansion of access beyond North America and Western allies, though the mechanism for clearing Japanese institutions reveals tighter government control than the public posture suggests.

The approval apparatus

Treasury Secretary Bessent is personally vetting and approving each company that gains access to Mythos. This level of direct government gatekeeping reflects administration concern about the model's capabilities—it can discover and exploit software security flaws faster than earlier technology—and the national security implications of misuse.

Japanese Prime Minister Shigeru Ishiba had instructed cabinet members to identify cybersecurity vulnerabilities in Japan's infrastructure and mitigate risks from advanced AI models like Mythos. Bessent's meeting with Ishiba appears to have surfaced the Claude access as a response to that directive.

The policy incoherence problem

The expansion to Japan sits awkwardly alongside simultaneous moves to restrict Chinese AI development. Jensen Huang is traveling to Beijing on Air Force One with Trump, in part to sell advanced AI chips that China will use to develop its own frontier models. A host of speakers flag this as policy contradiction: the administration is simultaneously approving access to Mythos for trusted allies while enabling China's chipmakers to close the gap.

The inference is unavoidable. If Mythos access is classified as sensitive enough to require Treasury vetting for each approver, the logic of unrestricted chip sales to Chinese hyperscalers becomes harder to square. One speaker notes: "The administration's AI policy remains inconsistent, incoherent. It is impossible to justify these two approaches simultaneously."

China's own reckoning

The near-term tension may be academic. Dario Amodei has stated that China's open-source AI community is roughly six months behind Mythos-level capability. When China achieves parity—something the transcript characterizes as likely within 12 months even on conservative timelines—the dynamics will shift. Unlike Anthropic, which has 15 co-founders, multiple funding rounds, and is likely to become a public company with SEC oversight, DeepSeek (the leading Chinese AI company) is closely held by its founder, who maintains majority voting control and invested substantial personal capital at low dilution.

That concentration matters. In the U.S., the "Mythos moment"—when a frontier model gets restricted or regulated—involved negotiation between Anthropic, the Department of Defense, and later the Treasury. Multiple stakeholders constrained unilateral action. In China, if DeepSeek achieves Mythos-level capability, the founder faces a very different counterparty: the CCP, which has a track record of asserting control over large tech companies through pressure on executives.

The question, still unanswered: what does a Chinese "Mythos moment" actually look like? The transcript suggests the stakes are higher and the precedent more opaque than anything the U.S. has yet negotiated.