OpenAI to supply Japan with cybersecurity AI under proposed government deal

OpenAI said on May 22 that it plans to provide the Japanese government and some Japanese companies with GPT-5.5-Cyber, a model the company says is designed for defensive cybersecurity work. Paul Nakasone, an OpenAI board member and former head of U.S. Cyber Command, told reporters in Tokyo that the company had held discussions with Japanese officials this week and wanted Japan to be among the first governments to receive the model. Jason Kwon, OpenAI’s chief strategy officer, separately told The Asahi Shimbun that the company also intends to broaden access to Japan and other allies. OpenAI launched GPT-5.5-Cyber in limited preview on May 7 for defenders responsible for securing critical infrastructure, saying the model is meant to support tasks such as vulnerability identification, malware analysis and patch validation while still blocking malicious uses. (japannews.yomiuri.co.jp) ### What exactly is Japan being offered? GPT-5.5-Cyber is OpenAI’s latest cyber-focused model, and The Yomiuri Shimbun reported that the company plans to provide access rights to the Japanese government and some companies. Nakasone said talks with Japanese officials included cybersecurity in 15 vital infrastructure fields, and the paper said the model can quickly detect weaknesses in critical infrastructure systems and help with remediation. (asahi.com) May 7 is when OpenAI formally rolled out the model in limited preview. In its own description, the company said GPT-5.5-Cyber is intended for specialized defensive workflows and is distributed through a vetting system called Trusted Access for Cyber, which lowers refusals for approved defenders while continuing to block requests tied to credential theft, stealth, persistence or malware deployment. (japannews.yomiuri.co.jp) ### Why is Japan part of this rollout now? Tokyo was the site of OpenAI’s announcement this week, and Kwon told The Asahi Shimbun the company was trying to broaden access to Japan and other allies. He said concern had grown globally over misuse of advanced cyber-capable AI systems after Anthropic introduced Claude Mythos last month. The British government’s AI Security Institute, cited by both OpenAI and Japanese media reports, found that GPT-5.5 had reached a performance level comparable to Mythos on cyber evaluations. (openai.com) The Yomiuri Shimbun said OpenAI expects Japan to become one of the first governments to receive the newest model. ### Where does the math claim fit into this? May 20 is when OpenAI published a separate research post saying an internal general-purpose reasoning model had disproved a longstanding conjecture tied to Paul Erdős’s 1946 unit distance problem. (asahi.com) OpenAI said the proof had been checked by external mathematicians and described the result as the first time an AI system had autonomously solved a prominent open problem central to a mathematical subfield. (japannews.yomiuri.co.jp) Tim Gowers, a Fields Medalist cited in OpenAI’s post, called the result “a milestone in AI mathematics,” while number theorist Arul Shankar said the paper showed current AI models were capable of “original ingenious ideas.” The company said the model was general-purpose rather than specialized for mathematics or narrowly scaffolded for that problem. ### What does this have to do with OpenAI’s market positioning? (openai.com) May 20 is also when reports emerged that OpenAI could confidentially file IPO paperwork within days and target a September public debut. Forbes, citing the Wall Street Journal, reported that Goldman Sachs and Morgan Stanley were among the firms advising on draft prospectus materials and that the plans remained fluid. (openai.com) Those reports arrived as rivals pushed on infrastructure as well as models. CNBC reported on May 21 that Anthropic was in talks with Microsoft to use Maia AI chips, with no deal signed, as Anthropic sought more computing capacity for Claude. Microsoft said in April that Maia 200 offers more than 30% improved tokens per dollar than the latest silicon in its fleet and is running in Arizona and Iowa data centers. (forbes.com) ### What happens next in the cyber rollout? June 1, 2026 is the next operational deadline in OpenAI’s cyber program. OpenAI said individual members of Trusted Access for Cyber who use its most cyber-capable and permissive models will be required to enable Advanced Account Security starting that day. Japan’s government access and any company deployments have not yet been given a public launch date. (cnbc.com) Nakasone said talks to move quickly on providing GPT-5.5-Cyber would continue, and Kwon said OpenAI plans to widen access in Japan as the company expands the program beyond the U.S. organizations that began receiving the model this month. (japannews.yomiuri.co.jp) (openai.com)