Blab AI
AI-Based Drug-Monitoring System Failed to Flag Fentanyl Theft at Tennessee Hospital
A nurse at Erlanger Medical Center in Chattanooga was fired in 2025 after stealing fentanyl from surgical waste, and a state nursing board order shows that the hospital’s AI‑powered drug‑monitoring software, Sentri7, did not raise alarms for months. The incident highlights a gap in the oversight of artificial‑intelligence tools that hospitals use to detect drug diversion.
The Tennessee Board of Nursing consent order, released in December 2026, documents that anesthesia staff noticed a nurse—later identified as John Stevenson—slurring his words and appearing exhausted while on duty in the surgery center around June 30, 2025. Stevenson admitted to diverting unused fentanyl that would otherwise have been discarded after surgeries, beginning in March 2025 and increasing to daily use by June. The order states that the nurse failed a drug test and was terminated.
Erlanger had implemented Sentri7, a medication‑monitoring platform developed by Wolters Kluwer. Sentri7 aggregates data from electronic health records and pharmacy systems to identify missing drugs in real time. According to the board order, the system failed to flag five instances of missing fentanyl and other inconsistencies in dispensing records that should have triggered alerts. The order notes that the software was in its “initial learning phase” at Erlanger, though it does not explain how that status affected performance.
Sentri7 is one of two AI‑driven drug‑diversion products that dominate the U.S. market, the other being Bluesight’s ControlCheck. Both platforms claim to detect diversion faster than manual methods. A 2022 NIH‑funded study found that Sentri7 could identify diversion cases up to a year earlier than human reviewers, but the study involved data from 10 hospitals and 22 known diversers.
Experts say the failure at Erlanger raises questions about the transparency and reliability of AI monitoring. David Rastall, a neurologist and AI researcher at Johns Hopkins Medicine, said that proprietary software and limited understanding by hospital staff can allow errors to go uncorrected. Jacob Smith, a pharmacist at Johns Hopkins, noted that the software is less effective in operating rooms, where drug dispensing and charting differ from emergency or intensive‑care settings.
The Tennessee Department of Health and the Tennessee Health Facilities Commission have no additional documents about the Sentri7 failure. Erlanger’s spokesperson, Charlie Milburn, declined to comment on the software’s performance or the incident. Wolters Kluwer’s executive Kristy Drollinger said in a March interview that Sentri7 is in high demand because many hospitals struggle to secure their drugs, but she did not discuss the Erlanger case.
Drug diversion is common in U.S. hospitals; the nonprofit Healthcare Diversion Network estimates that up to 15 % of healthcare workers divert drugs at least once. Diversion can lead to patient harm and has been linked to more than 200 infections, mostly hepatitis C, since 1985.
The incident is notable because hospitals are not required to disclose the use of AI monitoring systems or report software malfunctions. The Department of Justice’s Drug Enforcement Administration requires hospitals to report lost or stolen drugs, but those reports do not have to include details about AI tools. As a result, the prevalence of software failures remains largely unknown.
The board order concludes that Stevenson’s license was placed on probation and that he entered drug counseling. No criminal charges were filed. The case remains the first publicly documented instance of an AI drug‑diversion system failing to detect a theft in a U.S. hospital.
The situation underscores the need for clearer oversight of AI tools that safeguard medication safety. Hospitals, vendors, and regulators may need to establish reporting requirements for software performance and failures to prevent similar incidents in other facilities.
The Tennessee Board of Nursing consent order, released in December 2026, documents that anesthesia staff noticed a nurse—later identified as John Stevenson—slurring his words and appearing exhausted while on duty in the surgery center around June 30, 2025. Stevenson admitted to diverting unused fentanyl that would otherwise have been discarded after surgeries, beginning in March 2025 and increasing to daily use by June. The order states that the nurse failed a drug test and was terminated.
Erlanger had implemented Sentri7, a medication‑monitoring platform developed by Wolters Kluwer. Sentri7 aggregates data from electronic health records and pharmacy systems to identify missing drugs in real time. According to the board order, the system failed to flag five instances of missing fentanyl and other inconsistencies in dispensing records that should have triggered alerts. The order notes that the software was in its “initial learning phase” at Erlanger, though it does not explain how that status affected performance.
Sentri7 is one of two AI‑driven drug‑diversion products that dominate the U.S. market, the other being Bluesight’s ControlCheck. Both platforms claim to detect diversion faster than manual methods. A 2022 NIH‑funded study found that Sentri7 could identify diversion cases up to a year earlier than human reviewers, but the study involved data from 10 hospitals and 22 known diversers.
Experts say the failure at Erlanger raises questions about the transparency and reliability of AI monitoring. David Rastall, a neurologist and AI researcher at Johns Hopkins Medicine, said that proprietary software and limited understanding by hospital staff can allow errors to go uncorrected. Jacob Smith, a pharmacist at Johns Hopkins, noted that the software is less effective in operating rooms, where drug dispensing and charting differ from emergency or intensive‑care settings.
The Tennessee Department of Health and the Tennessee Health Facilities Commission have no additional documents about the Sentri7 failure. Erlanger’s spokesperson, Charlie Milburn, declined to comment on the software’s performance or the incident. Wolters Kluwer’s executive Kristy Drollinger said in a March interview that Sentri7 is in high demand because many hospitals struggle to secure their drugs, but she did not discuss the Erlanger case.
Drug diversion is common in U.S. hospitals; the nonprofit Healthcare Diversion Network estimates that up to 15 % of healthcare workers divert drugs at least once. Diversion can lead to patient harm and has been linked to more than 200 infections, mostly hepatitis C, since 1985.
The incident is notable because hospitals are not required to disclose the use of AI monitoring systems or report software malfunctions. The Department of Justice’s Drug Enforcement Administration requires hospitals to report lost or stolen drugs, but those reports do not have to include details about AI tools. As a result, the prevalence of software failures remains largely unknown.
The board order concludes that Stevenson’s license was placed on probation and that he entered drug counseling. No criminal charges were filed. The case remains the first publicly documented instance of an AI drug‑diversion system failing to detect a theft in a U.S. hospital.
The situation underscores the need for clearer oversight of AI tools that safeguard medication safety. Hospitals, vendors, and regulators may need to establish reporting requirements for software performance and failures to prevent similar incidents in other facilities.
