OpenAI has given the three largest banks in Japan access to its latest cybersecurity model GPT-5.5 Cyber, to help the organisations identify vulnerabilities and defend against advanced cyberattacks.

According to Nikkei, Mitsubishi UFJ Financial Group, Mizuho Financial Group, and Sumitomo Mitsui Financial Group will be granted access to the model, which OpenAI only makes available to trusted partners.

GPT-5.5 Cyber is OpenAI’s latest cybersecurity model, which the firm says can carry out autonomous penetration testing and code validation.

The launch of advanced cybersecurity models such as GPT-5.5 Cyber have raised concerns that hackers could soon leverage similar models – or hijack legitimate ones – to carry out more sophisticated attacks.

Recent analysis by the UK’s AI Security Institute found that GPT-5.5 Cyber is competitive with, and sometimes outperforms, Claude Mythos Preview. The model is only the second to complete the institute’s 32-step corporate network attack simulation, after Mythos Preview.

Earlier in May, Anthropic reportedly agreed to give the same three banks access to Claude Mythos Preview, its advanced cybersecurity model announced in April.

At the time Reuters reported that Japan would also establish a public-private working group to explore the potential cybersecurity impacts of Mythos Preview.

Anthropic has provided Mythos Preview to select organisations through Project Glasswing, a cross-industry cybersecurity initiative including companies such as Amazon Web Services, CrowdStrike, Google, JPMorgan Chase, Microsoft, NVIDIA, and Palo Alto Networks.

At the time of the announcement, no UK or European organisations were included.

Like GPT-5.5 Cyber, Claude Mythos Preview is intended to provide advanced vulnerability detection and other defensive cybersecurity capabilities.Anthropic has claimied the model found “thousands of high-severity vulnerabilities” in common operating systems in early tests.

Lack of access to Mythos Preview has since raised concerns among European financial institutions. This week, the European Central Bank called an urgent meeting with bank leaders to discuss measures they can take to protect their IT environments following the prospect of advanced AI-powered attacks raised by models such as Mythos Preview and GPT-5.5 Cyber.

---