What happened
According to a Tennessee Board of Nursing order reported by KFF Health News and CBS News, a nurse at Erlanger Baroness Hospital in Chattanooga diverted fentanyl over several months in 2025. KFF reports the diversion began in March 2025 with unused fentanyl that would otherwise have been wasted after surgical procedures, escalating from once or twice a week to daily use by June. Coworkers reported that he appeared impaired on June 30, 2025, which is what ended the diversion.
Where the AI fell short
The hospital uses Sentri7, an AI drug-diversion monitoring tool from Wolters Kluwer. KFF reports that a hospital diversion specialist, reviewing dispensing records after the fact, found about five instances of missing fentanyl waste that Sentri7 had not flagged, and that the software was in its initial learning phase at the time. KFF says Erlanger declined to comment on Sentri7, while a Wolters Kluwer spokesperson declined detailed questions but said the company remains confident in its software.
Why it matters
- •KFF notes hospitals are not required to disclose use of these tools or to report malfunctions, and a Johns Hopkins neurologist and AI researcher told KFF that limited transparency can let errors go unexamined rather than fixed.
- •Generic industry view: rare-event anomaly detection is hard, and model performance during early learning periods, plus data-quality gaps in waste logging, are common failure modes that argue for independent validation and sustained human oversight.
Key Points
- 1Per a Tennessee Board of Nursing order reported by KFF Health News, human coworkers, not the AI system, caught the diversion after the nurse appeared impaired on June 30, 2025.
- 2A hospital specialist later found about five instances of missing fentanyl waste that Sentri7 did not flag; KFF reports the software was in its initial learning phase at the time.
- 3Industry context: hospitals are not required to disclose use of such tools or report malfunctions, and observers say opacity can let anomaly-detection failures go unexamined.
Scoring Rationale
A documented, well-reported failure of a widely used AI drug-diversion tool at a major hospital is a notable AI-reliability and healthcare-safety story with real governance and transparency implications for ML practitioners. It is meaningful but localized and partly attributable to an early learning phase, keeping it in the notable band. Adjusted slightly from 6.8.
Practice with real Health & Insurance data
90 SQL & Python problems · 15 industry datasets
250 free problems · No credit card
See all Health & Insurance problems
