Diversion of controlled substances drains money from US hospitals and fuels addiction among staff. AI software now scans for it, and a STAT report published this week argues the tools are strong detectors that still fail when humans ignore them. Training and trust in the alerts are the missing pieces.
STAT opens with a 2024 case at Adventist Health Bakersfield. Family members saw a nurse wander the ICU barefoot, muttering, and handle IV needles carelessly. They later described those scenes to federal investigators.
Tools like ControlCheck and BD’s Sentri7 analyze dispensing records, waste logs and access patterns to flag nurses and pharmacists who handle controlled substances in ways that deviate from normal practice. Hospitals that pair the software with clear escalation paths catch diversion earlier, the report says. Facilities where alerts go to busy managers or skeptical supervisors see flags ignored, letting theft continue.
The human step matters because AI alerts are probabilistic, not proof. A flagged pattern can reflect sloppy documentation rather than theft, so staff need training to triage findings without wrongly accusing colleagues. When the loop closes, hospitals report catching diversion that had gone undetected for months.
For health systems, the lesson is that surveillance software is only half the answer. The other half is workflow: who reviews the alerts, how quickly, and whether clinicians believe the system is there to help patients rather than punish staff.
