Japan's Three Megabanks Will Get Anthropic's Mythos Model

Japan's three largest banks (MUFG, SMBC, and Mizuho) are set to gain access to Anthropic's Claude Mythos within roughly two weeks, according to Nikkei Asia and corroborated by TheNextWeb and Nippon.com. They will be the first Japanese institutions added to Project Glasswing, Anthropic's restricted rollout of the model.

Mythos is the model that can identify and, where permitted, exploit software security vulnerabilities significantly faster than previous systems. That capability is precisely why it is gated. Under Glasswing terms, partners use Mythos to find weaknesses in their own infrastructure and to draft remediation. They are not allowed to publish exploits or operate against third parties.

A government-coordinated patching push

The structural part of the announcement is more interesting than the bank-level deployment. Finance Minister Satsuki Katayama said the government is setting up a 36-entity public-private working group on Mythos-class risks. The membership includes the three megabanks, the Bank of Japan, and the Japanese units of Anthropic and OpenAI. Mizuho's chief information security officer chairs the group, which is charged with identifying exposures, implementing defensive measures, and drafting contingency plans for what amounts to a coordinated patching push across the Japanese financial system.

U.S. Treasury Secretary Scott Bessent was reportedly in Tokyo this week and informed the megabanks of the move in person. That detail says something about how Washington is positioning Mythos access. It looks more like a diplomatic asset than a pure commercial product.

What is unusual about the framing

The deployment inverts the usual safety conversation around frontier models. A system dangerous enough to require restricted access is being released specifically because attackers can build something like it sooner or later, and defenders need to close the gap first. That argument has been made in offensive-security circles for years, but rarely with state-level coordination on the receiving end.

This is part of a broader Anthropic enterprise push that includes the recent KPMG global alliance and the $15 billion compute deal with SpaceX. The Mythos-Japan arrangement appears to be the first time a foreign government has organised a national working group around a single AI model. It probably will not be the last.