Anthropic Project Glasswing Targets Safer AI Agents Before They Go Rogue
…Cloudflare turned Mythos Preview loose on its critical systems and uncovered roughly 2,000 bugs, 400 of them high or critical severity, with a false-positive rate that beat human penetration testers…