Anthropic CEO sets a half-to-one-year window for Mythos technical barriers

On April 18, AnthropicCEO DarioAmodi said in an exclusive interview with the Financial Times "LunchwiththeFT" that he speculated that open source models and China developers could replicate ClaudeMythos's cybersecurity capabilities within 6 to 12 months. Mythos is the model with the highest capability limit of Anthropic and has not been publicly released. It has been internally determined as a national security risk due to its ability to discover large-scale zero-day software vulnerabilities. In the past two weeks, the White House has pushed for access to federal agencies, and U.S. officials have urgently held discussions with major banks to urge them to use Mythos to strengthen their cybersecurity. Amodei also revealed that Anthropic has partnered with more than 40 organizations to promote the Project Glasswing vulnerability patching project, including Amazon, Apple and Microsoft. He admitted that the company's own data security practices were also under scrutiny because of the leaks of some Claude code. His prediction means Anthropic is doing two things at the same time: slowing the spread of Mythos-level capabilities while pushing the U.S. government and critical infrastructure to use it to fix system vulnerabilities first. The day the ability spreads to everyone, the foundation of defense must be in place.