Hospitals are turning into cognitive composite systems — humans, AI models, agents, devices, and infrastructure working as one. When the power goes out, the network drops, or the model is wrong, lives depend on what happens next. Fail Systems is a framework for understanding, classifying, and designing against that failure.
Failure in automated healthcare is not one thing. The Fail Systems framework classifies it into distinct layers — each with its own likelihood, blast radius, and defense.
Grid loss, UPS exhaustion, cooling failure, cloud-region outage. The layer everything else stands on.
Network partitions, EHR downtime, ransomware, corrupted or unavailable patient data at the moment of decision.
Monitors, pumps, ventilators, lab systems — firmware faults, update cascades, silent sensor drift.
Wrong outputs delivered confidently, agent loops acting on stale state, automation that fails without announcing it.
The seams. Skills that atrophied under automation, alarms nobody owns, the moment a human must take over and can't.
Small faults that propagate across layers — one bad update, one dead datacenter, one wrong default, system-wide.
A safe automated system is one that knows how to become a less automated system. Every clinical workflow should have a defined answer at each tier — before the failure, not during it.
AI and agents operate the workflow; humans supervise by exception.
Automation degraded or distrusted — humans decide, machines advise.
Electronics up, cognition down. Staff run the workflow on devices alone.
Power or network gone. Paper, batteries, hand calculation, human judgment — and a plan for it that has actually been rehearsed.
The full methodology: failure classes, severity scoring, and degradation-tier design for clinical systems. FMEA for the AI-hospital era.
Coming soonA structured self-assessment for hospitals and health-tech builders: where would your system break first, and what happens then?
Coming soonReal incidents — outages, update cascades, EHR downtime, AI failures — analyzed through the framework, built on the AI Med Risk incident library.
Coming soon