Overview
- OpenAI announced on May 21 that it will provide its specialized GPT-5.5-Cyber model to the Japanese government and a curated set of private partners to help find and patch software flaws.
- The model is being supplied to Japan’s three megabanks—MUFG, SMBC, and Mizuho—through a “Trusted Access for Cyber” program that vets users before they can run high-capability security scans.
- Anthropic’s Claude Mythos, which demonstrated autonomous zero-day discovery when it launched in April, is reported to be made available to the same megabanks by the end of May for defensive use.
- A 36-member public-private working group chaired by Mizuho’s chief information security officer was formed in mid-May to set guardrails, coordinate responses, and manage disclosure of newly found vulnerabilities.
- Regulators and global bodies warn that concentrating these powerful defensive tools risks misuse, creates a two-tier security landscape for smaller firms, and shortens the time defenders have to patch chained exploits.