JPMorgan Chase CEO Jamie Dimon said AI-related cyber risks had risen tenfold after Anthropic’s Mythos, in a statement reported on Oct. 6, 2026. The figure was Dimon’s assessment; a separate evaluation by the UK AI Security Institute (AISI) examined Claude Mythos Preview in controlled cyber exercises.
Jamie Dimon’s warning about Mythos
Dimon made the statement in a television interview about cybersecurity. His tenfold figure describes his assessment of the risk, while AISI’s evaluation examined a specific model in simulated exercises. They address different questions: one is a broad risk judgment, the other a set of controlled tests.
Anthropic said on April 7, 2026, that Claude Mythos Preview had found vulnerabilities across major operating systems and web browsers when directed to do so. The company also announced Project Glasswing, an initiative for limited access to Mythos Preview focused on defensive work.
What AISI evaluated
AISI evaluated Claude Mythos Preview in cyber ranges—simulated environments used to assess cybersecurity tasks. The institute said its ranges lacked active defenders and common defensive tools found in real environments. Those conditions describe the tests; they are not live corporate networks.
The AISI results concern Claude Mythos Preview, not every Anthropic model or a general measure of cyber risk. In particular, the evaluation does not provide a basis for treating Dimon’s tenfold assessment as a measured outcome.
Mythos Preview and Mythos 5.1 access
On Oct. 7, 2026, Anthropic’s Mythos page described access to Mythos 5.1 as limited to a small set of vetted organizations. That statement concerns Mythos 5.1; AISI’s cyber evaluation covered Claude Mythos Preview.
