Japan's three largest banks are set to gain access to Anthropic's latest artificial intelligence model, Claude Mithos, as early as this month.
The Nikkei reported Thursday that MUFG Bank, Sumitomo Mitsui Banking Corp. and Mizuho Bank appear to have received this information during a meeting the previous day with U.S. Treasury Secretary Scott Bessent, who was visiting Japan.
Anthropic declined to confirm the report, saying it was "not appropriate to comment on individual discussions." Amid growing concerns that Mithos, which has advanced reasoning capabilities, could be exploited for cyberattacks, Anthropic has limited access to the model and formed "Project Glasswing" with major tech companies including Google, Microsoft and Apple to build a joint response framework. Currently, only 40 to 50 entities, including U.S. government agencies, have access to Mithos, and outside the United States, access has reportedly been granted only to the U.K.'s AI Security Institute (AISI).
Since the "Mithos shock" in April, the Japanese government has asked financial institutions, electric utilities and other companies to inspect and address vulnerabilities in their information technology (IT) systems while seeking to secure access to Mithos. Japanese Prime Minister Sanae Takaichi also recently instructed ministers at a cabinet meeting to prepare for AI-related cyberattacks.
