Anthropic's Jack Clark: AI may be behind Japan cyberattacks

Clark's comments came at a press briefing marking the first anniversary of Anthropic Japan, the company's Japanese arm — not as a standalone security report. That setting helps explain why the message paired a warning with an offer of cooperation from the local unit.

Anthropic Japan representative Hidetoshi Tojo framed the defensive gap as one of speed. He said the current wave of attacks was "somewhat foreseeable," pointing to the Mythos model's arrival in spring 2026 as the point when such talk began. Tojo also said the Claude Mythos model is sometimes mistaken for "the only AI for defense," and argued that other models should be used in cyber defense as well so responses can move faster.

Clark's own framing stayed narrow: he described possible AI involvement as something whose details are unknown, and separately ruled out, as far as he knows, any case of AI attacking autonomously. The picture he offered is therefore one of hackers repurposing publicly available open-weight models, not of self-directed AI attackers — and Anthropic's answer, in his telling, is safeguards on model misuse plus continued work with Japanese government and corporate partners.