Artificial intelligence as adversary: Understanding Claude Mythos and the road ahead

On April 7, 2026, Anthropic disclosed that it had developed a model deemed too capable for unrestricted public release, Claude Mythos Preview (Mythos), a development with substantial implications for the legal, compliance and cybersecurity sectors. The gravity of the announcement was underscored almost immediately: the US Department of the Treasury and the Federal Reserve convened an emergency session with senior bank executives the very next day and the IMF has since warned that models of Mythos's caliber present material risks to global financial stability.

For now, access to Mythos remains tightly controlled, limited to roughly 40 vetted organizations participating in a program known as Project Glasswing. However, experts estimate that comparable capabilities could reach the wider market, including potentially hostile actors, within a window of six to 24 months.

Read our full analysis, “When AI becomes the cyber attacker: Mythos and what comes next" to understand our key takeaways.