According to a source familiar with the issue, MUFG, Mizuho, and SMFG will be the first Japanese institutions to be included in Anthropic’s limited Project Glasswing rollout, as reported by Reuters.
Japan’s three major banks will gain access to Claude Mythos, Anthropic’s AI model for finding vulnerabilities, in about two weeks, a source close to the matter informed Reuters on Tuesday.
This marks the first time a Japanese company will enter the restricted preview, previously limited to Anthropic’s U.S. and select European partners.
Mitsubishi UFJ Financial Group, Mizuho Financial Group, and Sumitomo Mitsui Financial Group were notified of this development during meetings in Tokyo this week with U.S. Treasury Secretary Scott Bessent. The three banks are anticipated to be onboarded by the end of May.
Mythos has been regarded by regulators and executives as a significant event since Anthropic revealed its existence earlier this month.
The model has identified thousands of previously unknown zero-day vulnerabilities across all major operating systems and web browsers, and in internal tests, it produced working exploits, including methods to bypass browser and operating-system sandboxes.
Last week, Mozilla released Firefox 150, addressing 271 vulnerabilities discovered by Mythos in a single evaluation.
Anthropic has not publicly released the model, opting instead for a controlled rollout under “Project Glasswing,” with 12 named launch partners such as AWS, Apple, Cisco, Google, JPMorganChase, Microsoft, Nvidia, and Palo Alto Networks, along with around 40 other institutions gaining access on a case-by-case basis.
The inclusion of Japan comes after the Federal Reserve and U.S. Treasury briefed American bank executives on cyber risks and after UK regulators committed to briefing major British banks within days.
Tokyo is acting in parallel. Finance Minister Satsuki Katayama announced the creation of a 36-entity public-private working group on Mythos-class risks, including the country’s major banks, the Bank of Japan, and the Japanese branches of Anthropic and OpenAI.
The group, led by Mizuho’s chief information security officer, will address exposures, develop defense measures, and plan for a coordinated patching effort across the Japanese financial system.
For the involved banks, the immediate concern is operational. Mythos, under Glasswing terms, limits output disclosure, using the model to detect a partner’s system vulnerabilities and draft solutions rather than publish exploits.
The Mozilla scenario offers a precedent: 271 vulnerabilities were fixed in one Firefox release following a Mythos evaluation, with findings shared with Mozilla engineers under non-disclosure instead of public release.
The geopolitical aspect is noticeably evident. Bessent’s role in communicating the access decision in Tokyo aligns Mythos rollout with U.S. Treasury policies instead of Anthropic’s commercial strategy, a move that has faced criticism from European leaders.
Eurozone finance ministers addressed the matter at an Ecofin meeting last week, where no EU government had access to the model, while the White House reportedly prevented further expansion of the partner list.
Opinions on Mythos are divided in the industry. Some cybersecurity experts argue that the vulnerabilities identified by Mythos are accessible through skillful use of public models, suggesting that the main focus should be on the rapid advancement of frontier AI in