Hospital AI Can "Drift" Without Anyone Noticing. Who Should Be Held Accountable?

Artificial intelligence systems deployed in hospitals for diagnosis, triage, or patient flow management don't remain static once they go live. They can experience a phenomenon known as "drift" — a gradual deviation in performance from initial baseline parameters, caused by changes in patient populations, clinical practices, or even how staff enter data. The problem, according to an opinion piece published by STAT News AI, is that very few medical institutions actively monitor this phenomenon.

Quarterly Governance Is No Longer Enough

The authors of the piece, physicians Peter Pronovost, Justin Norden, and Kedar Mate, argue that the current oversight model — in which a technical committee reviews AI system performance once every three months — has been outpaced by the actual speed of operational change. According to STAT News AI, governance of medical algorithms "must be treated as a daily responsibility" of hospital leadership, not as an occasional agenda item.

The central argument is that an algorithm initially validated on a particular dataset can begin producing less reliable results once real-world conditions shift — for instance, the emergence of a new viral variant, changes to treatment protocols, or even updates to the medical equipment feeding data into the system. Without constant monitoring, these deviations can go unnoticed for months, with potential consequences for clinical decisions based on the algorithms' recommendations.

A Responsibility for Executive Leaders

A key point in the opinion piece is that responsibility should shift toward hospital executive leadership, rather than resting solely with IT departments or specialized AI technical teams. The authors suggest that medical and administrative directors should build algorithm performance checks into their regular management routines, much as they already monitor other critical patient safety indicators.

The piece published by STAT News AI is part of a broader debate on regulating artificial intelligence in healthcare, as more medical institutions worldwide adopt such technologies without yet having clear standards for continuous oversight. The authors don't offer a single technical fix, but rather call for a shift in hospital organizational culture — one in which AI monitoring becomes part of leadership's daily work rather than a periodic administrative exercise.

For now, the opinion piece remains more of a warning signal than a practical guide, but it raises a question that an increasing number of healthcare systems will need to confront: who is actually responsible when a medical algorithm starts to gradually fail, and no one notices in time?