AI Red Team

A fake AI agent skill that passed multiple security scanners compromised 26,000 agents by using a mutable external link that pointed to benign code during review and malware after approval. The attack…