Overview
- OpenAI confirmed it has given selected Japanese financial institutions access to its GPT-5.5 Cyber model, Finance Minister Satsuki Katayama said on Friday after talks with OpenAI officials.
- The three biggest banks — MUFG, Sumitomo Mitsui Banking Corporation, and Mizuho — are the primary recipients reported to be onboarded for defensive use of GPT-5.5 through OpenAI’s Trusted Access for Cyber program.
- Anthropic’s Claude Mythos, which demonstrated the ability to autonomously find zero-day vulnerabilities when it was released in April, is also being provided to Japan’s megabanks and is expected to be available by the end of May.
- Japanese authorities have formed a 36-member public-private working group chaired by Mizuho’s CISO to set guardrails, coordinate testing, and limit misuse of models that can both find and expose unknown security flaws.
- Experts and regulators warn concentration of these tools at large banks could leave smaller firms exposed and compress patching timelines from years to months, creating systemic risks that require formal oversight and broader access planning.