Japan is in negotiations with U.S. artificial intelligence company Anthropic to gain access to its latest AI model, according to government sources on Tuesday, amid growing concerns about cyberattacks powered by advanced AI tools, News.Az reports, citing Kyodo.

The discussions come as the Claude Mythos model is reported to be highly capable of identifying vulnerabilities in web browsers and other software systems, raising fears that it could also be exploited for offensive cyber operations targeting a wide range of infrastructure.

Japanese authorities aim to use the model to strengthen national cyber defenses, particularly against threats believed to originate from China and Russia, while also gaining technical expertise in advanced AI development, the sources said.

Government officials and representatives from Anthropic are expected to hold talks in the near future, with Japan preparing the necessary infrastructure to potentially deploy Mythos if access is approved.

Access to the Mythos model, which was introduced in April, is currently limited to certain technology companies such as Google LLC and select financial institutions. However, governments and public sector organizations are increasingly seeking permission to use it.

Experts have raised concerns that if misused, Mythos could enable cyberattacks capable of disrupting systems across multiple sectors and critical services.

With increasingly sophisticated AI systems expected to emerge, Japanese officials believe securing access is in the national interest, according to the sources.

Finance ministers and central bank governors from the Group of Seven advanced economies are also expected to discuss concerns over AI-driven cyber threats, including those linked to models like Mythos, during meetings in Paris next week.

In Japan, Chief Cabinet Secretary Minoru Kihara has emphasized that the government will “move quickly” to respond to potential misuse. Meanwhile, the Financial Services Agency has established a public–private framework aimed at strengthening cybersecurity countermeasures.

News.Az 

By Nijat Babayev