Anthropic’s AI Mythos can chain cyber exploits on its own

Mythos: An AI tool too powerful for public release discusses Anthropic’s most capable model to date, Claude Mythos Preview (aka Mythos), described as a “step change” in AI performance, particularly for cybersecurity tasks. The article notes Mythos was kept secret until a data leak revealed its existence and characterises it as both a powerful defensive tool and a potential serious offensive cyberweapon, leading Anthropic to sharply limit access.

According to Fortune, Mythos can autonomously find and chain software vulnerabilities at scale, raising concerns about how quickly such capabilities could enable exploits and domain takeovers, should they fall into the wrong hands. The piece states that access is limited to a small number of organisations worldwide, with the NSA reportedly using Mythos Preview to stress‑test and harden sensitive systems, despite the Pentagon labelling Anthropic as a supply chain risk.

It also explains that Mythos can discover vulnerabilities across large codebases faster and more reliably than existing tools, and can combine multiple flaws into multi‑step exploit chains. Published on 20 April 2026, the article frames the development as a global cybersecurity risk that warrants careful handling.