Schneier Pushes Back on Anthropic's "Project Glasswing" Secrecy
What Happened
In a Friday post, Bruce Schneier critiques Anthropic's Claude Mythos Preview — a model with advanced cybersecurity-vulnerability-finding capabilities restricted to 50 organizations under Project Glasswing. Schneier argues the restriction favors large tech vendors while freezing out academic researchers who have historically driven defensive security progress, calling for broader access and greater transparency.
My Take
Restricting offensive-capable AI sounds responsible and is probably counterproductive. Defenders are always outnumbered; their edge comes from collective knowledge flowing faster than attackers can adapt. Lock the tools behind a Fortune-500-only gate and you're helping well-resourced enterprises while leaving small businesses and open-source maintainers defenseless. The responsible-access debate is the AI governance fight of 2026 — watch for policy proposals that treat Project Glasswing as either a precedent or a cautionary tale.
Read Original Source