Four of five enterprises that secured AI agent identities still can't contain one that goes rogue

**Four of five enterprises locked their AI agent identities and still can't cage a rogue.** That's not a security gap—that's a trust collapse wearing a CISOs' disguise. Visa didn't just brag about defense; they aimed Anthropic's Mythos at their own payment rails, stitched minor flaws into live exploit chains, and then open-sourced the harness. That's what engineering depth looks like. Everyone else? They're holding a badge for a guard dog they can't feed. This matters because containment is the new perimeter. Identity is now a launchpad, not a lock. If 53% of firms have already watched an agentic incident go sideways, the remaining 47% are just early on the countdown. The signal here is raw: security is shifting from prevention to *response*—and from hoarding intel to sharing the hunt. Visa's open-source move is a dare: stop buying agility, start building it. Closer: You can give an agent a name, but if you can't kill its process, you've just styled the wolf you're feeding.