Cybersecurity briefing — 2026-07-23

Your AI agent's safety refusals were never the security boundary.