This episode of the Health Data Ethics podcast unpacks a strange behavior observed in Claude during Anthropic’s internal testing — specifically a simulated blackmail attempt — and why that kind of behavior matters for healthcare. What stood out most wasn’t the incident itself, but the way Anthropic handled it: they shared the whole incident, explained their tiered safety system, and outlined the steps they took to reduce the risk. That kind of transparency is rare. And it’s the kind of posture we need more of in healthcare AI — not just regulatory compliance, but thoughtful public communication when things get weird. Episode covers: – what we can borrow from Anthropic’s approach – parallels between tiered safety systems, cybersecurity, and clinical governance – why healthcare orgs need to plan for edge cases before rollout Would love to hear how others are thinking about this.