Japanese government institutions and banks are among the roughly 150 organizations across more than 15 countries that were given access to U.S. startup Anthropic’s new artificial intelligence model Claude Mythos on June 2.
Participating organizations must first meet Anthropic’s security requirements. Once approved, they will use the model to identify system weaknesses and bolster defenses against potential cyberattacks.
In response to the announcement, Finance Minister Satsuki Katayama confirmed that the Japanese government and several financial institutions are among those granted access.
“I am pleased that we can stand at the forefront of preparations,” Katayama said.
Advanced models such as Mythos, known as “frontier AI,” are exceedingly effective at uncovering system vulnerabilities.
This capability has raised concerns that, if misused, they could cause enormous damage to critical infrastructure such as financial systems.
For this reason, both Japan and the European Union have been pushing to secure access for defensive research.
The decision comes after Katayama revealed on May 22 that U.S. Treasury Secretary Scott Bessent had indicated access would be “granted within two weeks” during a visit to Japan.
In a related development, U.S.-based OpenAI is also expected to grant major Japanese financial institutions access to one of its frontier AI models.
The EU is also poised to gain access. According to sources, Anthropic has notified the bloc that it will grant access to the European Union Agency for Cybersecurity (ENISA).
At a news conference on June 1, European Commission spokesperson Thomas Regnier confirmed the commission has had several fruitful meetings with Anthropic, welcoming what he called the “latest developments” toward securing access.
(This article was written by Toshiki Horigome, and correspondents Ayumi Sugiyama and Kuniaki Nishio.)
