Report: NSA Is Already Using Anthropic's "Restricted" Mythos Model Despite Public Blacklist Posture
What Happened
Sherwood reported that the NSA has gained access to Anthropic's Mythos model under Project Glasswing, despite the model being formally restricted from broad release. The arrangement comes as Anthropic publicly maintains a high-bar disclosure regime, and arrives weeks after Anthropic and the federal government deepened their compute and security partnerships.
My Take
This is the predictable end-state of "responsible scaling policies" written by labs and audited by no one. If the model is safe enough for NSA's offensive teams, the argument it's too dangerous for vetted enterprise security customers collapses. The deeper governance question — who decides which agencies and which countries get frontier capability — is now urgent and currently has no answer outside Anthropic's San Francisco offices. Expect EU and UK regulators to formally raise this in May.
Read Original Source