When Japan's Finance Minister Satsuki Katayama confirmed on June 3 that the country's financial system would gain access to Claude Mythos, she was describing a deal that had been in motion for weeks. U.S. Treasury Secretary Scott Bessent had previewed the arrangement during a three-day visit to Tokyo in mid-May, framing it as part of a broader technology cooperation between Washington and Tokyo. By Tuesday, the details were official: Japan was joining the second wave of Project Glasswing participants, alongside roughly 150 other organizations in more than 15 countries.

The announcement came the same week Anthropic disclosed the broader Glasswing expansion to critical infrastructure sectors worldwide. Japan's inclusion stands out for its specificity. The three institutions confirmed so far are MUFG Bank, Mizuho Bank, and Sumitomo Mitsui Banking, three of the largest financial institutions in Asia by assets. The finance ministry itself also receives access, giving Japan's financial regulator direct visibility into what the model can find.

Japan's Claude Mythos Access: Key Details

  • Megabanks confirmedMUFG, Mizuho, SMBC
  • Government accessFinance Ministry (Katayama)
  • Wave 2 total organizations~150 in 15+ countries
  • Original Glasswing launchApril 7, 2026
  • Zero-days found by Mythos1,000+ critical-severity
  • Disclosure timeline90-day standard window

Why Banking Systems Are a Priority

Financial infrastructure is among the highest-value targets for state-sponsored and criminal threat actors. Core banking platforms, payment settlement systems, and securities clearing networks run on software stacks that are often decades old, maintained by small teams, and rarely subjected to the kind of systematic vulnerability scanning that large technology companies apply to their products. A model capable of finding critical-severity zero-days at the speed and scale Mythos has demonstrated is, for a finance ministry, exactly the kind of capability worth queuing for.

The framing from Katayama made the security rationale plain. "There is no doubt that this will contribute to strengthening security at a time when it is unacceptable for Japanese financial institutions to fall behind other countries in this field," she said. That framing, with its competitive edge, reflects a growing awareness in Tokyo that peer nations and allied financial systems are now scanning their own codebases with AI. Japan's inclusion in Glasswing means its institutions won't be starting that process from behind.

"There is no doubt that this will contribute to strengthening security at a time when it is unacceptable for Japanese financial institutions to fall behind other countries in this field." Finance Minister Satsuki Katayama, June 3, 2026

The Glasswing Structure

Claude Mythos Preview was designed from the outset to be deployed defensively rather than publicly. When Anthropic announced the model in April 2026, the company cited an internal finding that more than 99% of the vulnerabilities Mythos had discovered were still unpatched. Releasing the model broadly before defenders could address those exposures would, in Anthropic's assessment, benefit attackers more than defenders. Project Glasswing became the mechanism for using Mythos offensively on behalf of defenders, under controlled conditions and with standard 90-day responsible-disclosure timelines.

The first wave of Glasswing participants focused on operating-system vendors, browser makers, and cloud infrastructure providers. This second wave shifts the lens toward sectors where software vulnerabilities carry systemic risk: financial services, energy, water, and healthcare. Japan's megabanks sit squarely in that framing. A zero-day in core banking middleware, exploited at scale, is not merely a regulatory incident. It is a systemic event. Running Mythos across those stacks before an adversary does is, by that logic, risk management at the infrastructure level.

Diplomacy and Access

The diplomatic channel through which Japan gained access is also worth noting. Scott Bessent's Tokyo visit in May framed Claude Mythos access as part of U.S.-Japan technology cooperation, a signal that the U.S. government views Anthropic's security capabilities as relevant to alliance-level cyber defense. That framing puts Anthropic in a novel position: a private AI company whose most sensitive model is being discussed at the level of bilateral treasury diplomacy.

For Japan's financial sector, the practical questions are now operational. Project Glasswing participants receive Claude Mythos access under tight controls, with Anthropic providing structured support for triage and disclosure. Japanese institutions joining the program will need to coordinate findings across the three megabanks and the ministry, a process that has no clean precedent in Japan's financial oversight structure. The Finance Ministry's direct inclusion in the program, rather than a third-party intermediary, suggests Anthropic and the ministry have worked out at least the broad outlines of that coordination. How it works in practice will become clearer as Glasswing's findings begin arriving in Japanese financial codebases over the coming months.

Further reading: Learn more about Claude's model family, read our background on Anthropic, or browse the latest Claude AI news.