Japan seeks access to Anthropic AI amid cyberattack concerns

TOKYO — The Japanese government is negotiating with U.S. artificial intelligence company Anthropic to obtain access to its latest AI model, government sources said Tuesday, amid growing concerns over cyberattacks using sophisticated AI tools.

The move comes as the Claude Mythos model is said to be so good at identifying vulnerabilities in web browsers and other software that it has sparked fears it could be exploited for attacks on a wide range of systems.

Prime Minister Sanae Takaichi on the same day instructed the minister in charge of cybersecurity, Hisashi Matsumoto, to consider measures against cyberattacks using the latest AI tools.

Japan aims to use the model to strengthen its defenses against cyberattacks, including those believed to originate from China and Russia, while gaining knowledge and expertise related to AI development, the sources said.

Government officials and Anthropic are expected to hold talks soon, with Japan preparing the necessary environment to use Mythos if access is granted.

Matsumoto told a press conference after a Cabinet meeting that Japan is "exchanging views with the United States" over negotiations to obtain access to the model.

"At a stage where we do not have access to it, we must still move ahead with countermeasures," Matsumoto said.

The government plans to soon hold talks among relevant ministries and work with the ruling Liberal Democratic Party and others to draw up concrete countermeasures.

Access to the Mythos model, which was unveiled in April by Anthropic, is currently restricted to IT companies such as Google LLC and some financial institutions, but governments and public organizations are also seeking access.

Experts are concerned that if exploited, Mythos could be used to carry out cyberattacks capable of disrupting systems in a broad range of sectors.

Anthropic, cofounded in 2021 by former OpenAI executive Dario Amodei, is a U.S. startup known for developing generative AI with an emphasis on safety. It released the conversational AI Claude in 2023, adopting a system that trains AI based on principles that include avoiding harmful behavior and maintaining human oversight.

As sophisticated AI systems are expected to continue to emerge, the government believes it is in Japan's national interest to secure access, the sources said.

Finance ministers and central bankers from the Group of Seven advanced economies are expected to share their concerns about cyberattacks using the latest AI models, including Mythos, when they meet in Paris on Monday and Tuesday next week.

In Japan, Chief Cabinet Secretary Minoru Kihara has stressed that the government will "move quickly" in responding to possible misuse. The Financial Services Agency has set up a public-private framework to strengthen countermeasures.

© KYODO