Japan's megabanks get access to Claude Mythos to hunt for vulnerabilities

Japan's three largest banks, MUFG, Mizuho and SMFG, will get access to Claude Mythos, Anthropic's restricted AI model for finding vulnerabilities in code. They will be the first Japanese financial institutions in Project Glasswing, a controlled Anthropic pilot.

What Claude Mythos is

Claude Mythos is a specialized Anthropic model built specifically for security and vulnerability discovery in code. Its main function is automated scanning of source code for potential security problems, which lets companies find holes in their systems before attackers do.

This matters for financial institutions, where security is a question of reputation, the trust of millions of customers and compliance with strict regulatory requirements. Even a single compromised system can lead to leaked personal data, financial losses and lasting damage to a bank's reputation for years.

Project Glasswing: selective access

Project Glasswing is a limited Anthropic program in which Mythos is tested on a chosen group of organizations before a broader commercial launch. The Japanese megabanks get access roughly two weeks after the announcement, making them the first Asian financial institutions in the program.

The choice of these three banks is not accidental. They are not just large institutions but cornerstones of the Japanese financial system:

• MUFG (Mitsubishi UFJ Financial Group): Japan's largest bank by assets, with more than 3 trillion dollars under management
• Mizuho Financial Group: the second-largest financial conglomerate, with a history of 400+ years
• SMFG (Sumitomo Mitsui Financial Group): the third player in the trio, serving large corporations and government projects

Together, these banks process trillions of yen a day and require the highest level of cybersecurity. Picking them as pilot partners reflects Anthropic's recognition of the high security standards in Japan's financial sector.

Why banks need this

The financial sector is under constant attack from cybercriminals, hackers and state actors. Vulnerabilities in systems are not just technical glitches; they are potential holes through which customers' money can leave and an institution's reputation can be compromised.

Claude Mythos lets banks scale security audits without hiring huge teams of pentesters for every system. Automated vulnerability discovery is especially useful for legacy code: old systems that every large bank has but rarely updates, for fear of disrupting working processes.

A specialized model trained specifically on vulnerability-finding tasks delivers higher accuracy than general-purpose solutions. That means fewer false positives and more useful information for security engineers, who can focus on real problems.

What this means

The Japanese megabanks' decision to use Claude Mythos signals that specialized AI security models are moving from an expert niche into mass use by large corporations. The financial sector is known for its conservatism in adopting new technology, so the choice of Japan's largest banks carries symbolic weight. It boosts confidence in the product and creates momentum for other financial institutions worldwide.

Other megabanks in the US, Europe and Asia may soon ask for access to similar tools.