Anthropic loosens secrecy around Claude Mythos Preview after criticism

Anthropic's Claude Mythos Preview, described as the scariest AI model in the world due to its cybersecurity capabilities, has been shrouded in secrecy. Initially, only about 50 companies and organizations were granted access through Project Glasswing, and all participants were required to sign confidentiality agreements to prevent findings from being shared. The Wall Street Journal reported that this changed after Democratic Representative Josh Gottheimer wrote a letter to Anthropic, arguing that no entity should be contractually restricted from warning others about urgent cyber risks. Anthropic responded by evolving the user agreements to ensure key information can be shared broadly, even beyond Project Glasswing’s bounds.

This shift comes as OpenAI launched a similar program called Daybreak, which was far less secretive from the start, allowing anyone to fill out a brief form and request a codebase scan. Already, companies like Cloudflare have begun speaking publicly about Mythos Preview. Grant Bourzikas, Cloudflare’s CSO, published a detailed post describing how Mythos can take low-severity bugs that traditionally sit invisible in a backlog and chain them into a single, more severe exploit. He also invited other teams to compare notes, signaling a new era of transparency. While the mystique around Claude Mythos Preview may fade, the move toward openness could accelerate cybersecurity collaboration.