OpenAI has granted some of Japan's largest banks access to its newest GPT-5.5 model for cyber defense, Finance Minister Satsuki Katayama said Friday.
Key Points:
- OpenAI has opened access to its GPT-5.5 model for some Japanese banks to help blunt cyberattacks, the finance minister said Friday.
- MUFG, Sumitomo Mitsui and Mizuho are the expected early users, reached through a vetted channel OpenAI calls Trusted Access for Cyber.
- The step parallels Japan's separate plan to deploy Anthropic's Claude Mythos, hinting at an emerging market for state-backed AI defense.
OpenAI Hands Japan's Megabanks GPT-5.5
Katayama said the access would help lenders respond to a fast-rising wave of attacks, speaking after a Tokyo meeting with OpenAI chief strategy officer Jason Kwon.
The release of AI that can write high-level code has handed hackers an unusual ability to find weak points and exploit them faster. She called early access one of the few real defenses, and did not name the institutions involved.
Local reporting named MUFG Bank, Sumitomo Mitsui Banking Corp and Mizuho Bank as the expected first users, with the model reserved for trusted partners and rated on par with Anthropic's rival system. OpenAI is delivering the cyber variant through a vetted program it calls Trusted Access for Cyber, which keeps the most capable tools with verified defenders.
Also Read: Kalshi Wins CFTC Approval For First U.S. Bitcoin Perpetual Futures
Officials Frame GPT-5.5 As Financial Defense
For the banks, the immediate goal is operational. Officials want the model pointed at their own systems, to spot intrusions and weak points before attackers can reach them first. The early effect is narrow but real: three of the world's largest lenders will soon have a frontier model aimed at their own defenses.
The logic is gatekeeping. A model strong enough to find software flaws at scale is, by the same measure, dangerous in the wrong hands, so access is rationed to institutions that can be vetted.
The deal reached beyond the technical level, with US Treasury Secretary Scott Bessent helping broker terms that read like a government-to-government understanding.
Both OpenAI and Anthropic are now courting sovereign customers with cyber-specific versions of their flagship models. That marks the early shape of an AI defense-contractor market, with banks and finance ministries as the buyers.
There is a structural catch inside the good news. Concentrating the strongest defensive AI in a few large, vettable institutions leaves smaller banks and fintech startups on the far side of a widening gap. A two-tier landscape, where the megabanks are well guarded and everyone else is more exposed, is a plausible byproduct.
Claude Mythos Drove Japan's Cyber Push
The push did not start with OpenAI.
Anthropic disclosed Claude Mythos in April, a model that surfaced thousands of zero-day flaws across major operating systems and browsers and wrote working exploits in testing. Japan answered in mid-May with a public-private working group on the risks, and its government and banks are set to use Mythos for defense, weeks after OpenAI opened comparable access to European firms.
Read Next: Dogecoin Reserves Edge Up To 28B As Whale Support Stays Weak