OpenAI to Bring Cyber AI to Japan, Work with AISI

2026-05-29

Cyber defenses strengthened

It became clear on the 29th that US-based OpenAI has provided access in Japan to its latest artificial intelligence (AI). Following US-based Anthropic, the move highlights how public and private sectors in Japan and the US are stepping up preparations against cyberattacks. Both companies have been holding talks with allied governments in quick succession, underscoring how cyber defense is becoming the main battleground in the AI race.

On the same day, OpenAI announced the Japan Cyber Action Plan, an initiative to strengthen cyber defenses. As part of the plan, it will provide Japan with access to GPT-5.5-Cyber, which it unveiled on the 7th. Access will first be opened to three banks - Mitsubishi UFJ Financial Group, Sumitomo Mitsui Financial Group and Mizuho Financial Group - before expanding beyond financial institutions.

First Asia collaboration with AISI

OpenAI signed a memorandum of cooperation (MOC) with the AI Safety Institute (AISI), the government body that evaluates AI safety in Japan. For OpenAI, cooperation with a body of this kind follows the US and Britain, and is its first in Asia. It will make it possible to test risks in new models, among other things.

OpenAI Chief Strategy Officer Jason Kwon visited Japan and met on the 29th with Finance Minister Satsuki Katayama and Science and Technology Minister Kimi Onoda. After the talks, Katayama told reporters that OpenAI had promised to provide access to some Japanese financial institutions. 'This is welcome from the perspective of strengthening cybersecurity at Japanese financial institutions,' she said.

Kwon told Nikkei and other media outlets, 'We will build cutting-edge cybersecurity in Japan. We will expand priority and rapid access.'

Competition over performance intensifies

OpenAI's new AI model, GPT-5.5, is intended for general use and is publicly available. GPT-5.5-Cyber, whose provision to Japan was disclosed this time, is specialized for cyber use and can support a wider range of related measures.

Attention to high-performance AI was heightened by Anthropic's new AI model Claude Mytus, announced in April. While its ability to find vulnerabilities is strong, concerns were also raised that it could be misused for cyberattacks, and access was limited to about 50 companies and organizations, mainly US firms, as well as some US government agencies. Interest later broadened after governments around the world sought access.

OpenAI, which was behind in cyber-focused models, is moving quickly to catch up with GPT-5.5-Cyber. It is in talks with governments including Japan's, aiming to set the standard for cyber AI.

The UK government's AI Security Institute (AISI) tested the cyberattack capabilities of both companies' models. In a report released on April 30, GPT-5.5 was said to have roughly the same level of capability as Mytus, but a release on May 14 said Mytus had slightly surpassed it, showing that the assessment has yet to settle. US-based Google matches OpenAI and Anthropic in overall AI model performance, but in cyber-focused models, Mytus and GPT-5.5-Cyber are seen as the two leaders for now.

Allied defense ties expand

On the 15th, a senior Anthropic executive in charge of international policy also visited Japan and held talks with Liberal Democratic Party officials and others over access to Mytus. Behind the two companies' rapid moves toward Japan with their high-performance AI lies a desire to expand cyber defense cooperation among allied countries such as Japan before Chinese technology catches up.

In a benchmark Anthropic used to explain Mytus's performance, which measures AI vulnerability-handling capability, models from Chinese firms such as MiniMax Group's Haici Technology and venture Beijing Zhipu Huazhang Technology, spun out of Tsinghua University and known as Zhipu AI, also rank near the top.

OpenAI is expected to file for a public listing soon, and it appears to aim to draw investor interest by strengthening its presence in cybersecurity as well. Anthropic, meanwhile, said on the 28th that it will make technology with performance close to Mytus generally available within the next few weeks. It said the release is possible thanks to progress in technical verification and the development of safeguards to prevent misuse in cyberattacks.

The detection and exploitation of vulnerabilities by AI that surpasses humans could shake social infrastructure and national security. The development of AI for cyber defense is also becoming a flashpoint in the US-China struggle for supremacy.

Enjoyed this article? Share it with your network!