In the spring and early summer of 2025, a certified registered nurse anesthetist named John Stevenson was working at Erlanger Baroness Hospital in Chattanooga, Tennessee — and diverting leftover surgical fentanyl for his own use. He did it at least five times over roughly four months. The hospital had installed Sentri7, an AI-powered drug-diversion monitoring system made by Wolters Kluwer, specifically to detect this kind of behavior. It missed every instance. The system, according to hospital records reviewed by local and national media, was still in its initial learning phase — calibrating its baseline for what normal medication handling looks like before it could flag anomalies.
On June 30, 2025, coworkers reported Stevenson to hospital administration. The complaint was not about a spreadsheet anomaly or a flagged transaction log. It was about his physical condition on the unit: slurred speech, visible swaying, nodding off while standing — the classic presentation of opioid intoxication on duty. He failed a drug test the same day and was terminated from his position.
The Tennessee Board of Nursing reached a consent agreement with Stevenson in November 2025. Under the settlement, his nursing license was placed on probation with mandatory drug counseling and ongoing monitoring. No criminal charges were filed. He is not listed as currently practicing at Erlanger, and the hospital's spokesperson declined to comment on whether he had sought employment elsewhere. The outcome is not unusual — BON settlements for first-time diversion cases frequently end in probation, rehabilitation, and continued licensure rather than permanent revocation — but it has drawn sharp criticism in the nursing and anesthesia communities given the role that impaired anesthesia providers can play in patient harm.
The AI problem nobody wants to talk about
The Sentri7 system failure is the part of this story that deserves serious scrutiny. Hospitals have poured significant resources into AI-based drug-diversion monitoring platforms over the past decade, selling administration on the promise that algorithmic oversight can catch what human supervisors miss. The core claim is reasonable in theory: analyze wasting records, compare controlled substance volumes dispensed to volumes administered, flag statistical outliers, and surface cases for human review. The limitation — rarely discussed in vendor marketing materials — is that these systems must be trained on a baseline of what normal looks like at your specific facility before they can identify what abnormal looks like. During that learning window, they are largely blind.
Erlanger has not publicly confirmed how long the Sentri7 learning phase lasted or why the flagging thresholds had not been calibrated sufficiently to detect repeated diversion over a four-month window. What is confirmed is that the five incidents Stevenson is documented to have committed during that period did not generate any automated alerts that reached an administrator. The detection that ultimately ended the situation was entirely human — bedside colleagues who noticed something was wrong with their colleague and reported it through the hospital's HR chain. The AI did not save anyone. The nurses did.
I've worked alongside CRNAs in ICU and surgical settings for over a decade. Drug diversion at that level — by someone trusted with sole-provider anesthesia, often with minimal direct supervision during a case — is exactly what these AI monitoring systems are supposed to catch. The Sentri7 learning-phase failure raises a question hospitals aren't asking loudly enough: what is your diversion detection protocol when the AI is still calibrating? If the answer is "the same AI," you have a gap. The bedside staff who finally reported Stevenson's impairment were the last line of defense. They shouldn't have been the first. And the fact that no criminal charges were filed for a healthcare worker diverting opioids from surgical patients — patients who needed that fentanyl for pain management after going under — is a separate conversation the profession needs to have about accountability at the CRNA level.
For nurses and hospital staff who observe a colleague showing signs of impairment — slurred speech, confusion, coordination problems, unusual drowsiness — the appropriate action is to report to a charge nurse or supervisor immediately without confronting the individual directly. Most state boards of nursing operate peer assistance programs designed to support healthcare workers with substance use disorders while ensuring patient safety. In Tennessee, that program is the Tennessee Professional Assistance Program (TNPAP). Reporting a colleague is not a career-ending act; in most jurisdictions it is a protected action.