Schneier and David Lie Publish a Sharp Critique of Anthropic's Restricted "Mythos" Release
What Happened
In an essay co-written with University of Toronto's David Lie, Bruce Schneier argued that Anthropic's decision to limit Claude Mythos Preview to ~50 vetted organizations under "Project Glasswing" is corporate gatekeeping dressed as safety. The piece notes security firm Aisle replicated Mythos-style vulnerability findings using cheaper public models, and calls for "greater transparency and information sharing with the broader community."
My Take
This is a much sharper question than it looks. If Mythos-class capability is reproducible with public models — and Aisle's work suggests it is — then "restricted access" is theater that mostly serves Anthropic's commercial position with US government customers. The right framework isn't open-vs-closed, it's whether vulnerability-discovery capability gets disclosed responsibly to defenders before attackers reproduce it. Expect a congressional hearing on AI cyber-capability disclosure within six months.
Read Original Source