Japan’s three largest banks are set to gain access within weeks to Claude Mythos, the restricted artificial intelligence model developed by Anthropic, marking the first official deployment of the cybersecurity-focused system by Japanese companies as Tokyo steps up preparations against AI-driven attacks on critical financial infrastructure.

According to Nikkei, Mitsubishi UFJ Financial Group, Mizuho Financial Group and Sumitomo Mitsui Financial Group are expected to receive access to Mythos by the end of May after discussions involving US Treasury secretary Scott Bessent during his visit to Japan this week. Reuters cited a person with direct knowledge of the matter as saying the banks could gain access “in about two weeks”.

The paper reported that the banks are expected to complete security vetting and sign agreements with Anthropic before integrating the model into production environments. The system is designed to identify and exploit software vulnerabilities at speeds beyond earlier AI tools, enabling banks to detect weaknesses before attackers can weaponise them. Anthropic has restricted access to the model because of concerns that it could destabilise financial infrastructure if misused.

Japanese prime minister Sanae Takaichi told cabinet ministers on Tuesday to strengthen preparedness against AI-driven cyberattacks targeting critical infrastructure, including banks and utilities, according to Nikkei. The publication reported that access to Mythos was likely discussed during her meeting with Bessent.

Reuters reported that Japanese finance minister Satsuki Katayama said after meeting Bessent that Japan would establish a public-private working group this week to address cybersecurity risks linked to Mythos. Japan’s Financial Services Agency is expected to hold the group’s first meeting on Thursday alongside the Bank of Japan and the country’s megabanks.

Kyodo News reported that Anthropic said its Mythos preview “has already found thousands of high-severity vulnerabilities”. The company has limited access under its defensive-access programme, Project Glasswing, which already includes US technology groups such as Apple and Microsoft. Anthropic declined to comment on specific discussions with the Japanese banks, while the three lenders declined to comment to Reuters.

Reuters separately reported that Frank Elderson, vice-chair of the European Central Bank’s supervisory arm, warned euro area lenders on Wednesday to accelerate preparations for cyberattacks enabled by Mythos or similar AI systems. “Lack of access is not an excuse for inaction,” Elderson said in an interview published by the ECB. “On the contrary, it makes it even more critical that banks step up and act now.”

---