OpenAI’s GPT-5.5-Cyber heads to Japan’s largest banks

Japan’s three largest banks will soon get access to OpenAI’s newest cyber defence model, GPT-5.5-Cyber, according to Finance Minister Satsuki Katayama. The banks named are MUFG Bank, Sumitomo Mitsui Banking Corporation and Mizuho Bank.

The access will come through OpenAI’s “Trusted Access for Cyber” programme, which the company uses to limit the most capable models to verified defenders. The idea is to keep powerful vulnerability-finding systems in the hands of vetted institutions rather than open them up broadly.

Katayama said the arrangement was discussed with US Treasury Secretary Scott Bessent as part of the process that opened the collaboration. That gives the deal a government-to-government feel as well as a commercial one, with Tokyo treating cyber defence as a strategic capability.

The move fits into a wider Japanese push on AI-related cyber risk. In mid-May, Japan set up a public-private working group that brings together the major banks, the Bank of Japan and local units of leading AI labs, according to the source article.

That group was formed around concerns about a new class of vulnerability-hunting systems. The article says the best-known example discussed in that context is Anthropic’s Claude Mythos, and that Japanese institutions are separately set to gain access to it as well.

OpenAI’s deal makes it the second major US AI lab to join the same defensive effort in Japan. According to the article, both OpenAI and Anthropic are positioning cyber-specific versions of their flagship models as tools for national defenders.

That has turned Japan into an early test case for what the article describes as an AI defence-contractor market. In that setup, banks and finance ministries are the customers, while AI labs supply models tuned for security work.

The emphasis on access control is not accidental. A model that can scan systems for weaknesses at scale can also be dangerous in the wrong hands, so the programme is built around verification and gatekeeping.

The arrangement also reflects how governments are thinking about frontier AI. Rather than treating these systems only as consumer products or general-purpose software, Japan is treating them more like critical infrastructure tools that may need special handling.

The source article also flags a downside. By concentrating the most capable defensive AI in a small number of large institutions, the programme could widen the gap between the biggest banks and everyone else, including smaller banks and fintech startups.

For now, though, the immediate result is simple. Three of Japan’s megabanks will soon have access to GPT-5.5-Cyber through a vetted channel, with the collaboration helped along by officials in both Japan and the United States.