ANTHROPIC BUILT AN AI SO DANGEROUS IT REFUSES TO RELEASE IT — CLAUDE MYTHOS FOUND THOUSANDS OF ZERO-DAYS ON ITS OWN Anthropic’s Claude Mythos Preview autonomously discovered and wrote working exploits for thousands of critical zero-day vulnerabilities across every major OS and browser — capabilities so dangerous Anthropic decided not to release the model publicly. Instead, the company launched Project Glasswing: $100 million in usage credits and restricted access for vetted security researchers only. Keywords: Anthropic Claude Mythos, AI hacking, zero-day exploits, AI cybersecurity risks