Anthropic, the AI safety company born from the halls of OpenAI, has disclosed a fourth security breach — one that its own initial review failed to surface. The quiet emergence of this overlooked incident invites a deeper reckoning: in an era when AI systems are becoming load-bearing pillars of critical infrastructure, the institutions entrusted to build them safely must first prove they can see themselves clearly. Transparency, it turns out, is not a single act but a discipline — and the gap between what a company believes it knows and what is actually true can be consequential.