AI safety divergence: OpenAI enters the defense line of Japanese banks, while the Anthropic model is still subject to controlled evaluation.

Japanese Finance Minister Satsuki Katayama said that some Japanese financial institutions have been granted access to OpenAI's GPT-5.5 model for preventing cyber attacks. According to media reports, Katayama told reporters in Tokyo after meeting with OpenAI's Chief Strategic Officer Jason Kwon, "This is a promising development and an important step for Japanese financial institutions to strengthen their network defense capabilities." According to the Nikkei newspaper on Thursday, Japan's three largest banks - MUFG Bank, Sumitomo Mitsui Banking, and Mizuho Bank - are expected to access the latest model from OpenAI, which is backed by Microsoft. The model is currently only open to trusted partners and is believed to be comparable to Anthropic's Claude Mythos. After Anthropic launched its most powerful general-purpose model, Claude Mythos Preview, the EU is also in talks with the American AI giant to access its new network security model. Anthropic claims that this model is their "strongest version yet in terms of encoding and intelligence tasks," as Mythos is able to discover and exploit security vulnerabilities in IT systems, causing concern worldwide. Currently, the application of Anthropic's Claude Mythos model leans more towards security assessment and capability verification, mainly opening up to some partners in a controlled environment to test its potential application boundaries in network security scenarios.