Automatic operation timestamps as accurate as manual ones across seven surgical specialties, according to Aimbient

A Stanford Medicine-led study published in JAMA Surgery finds that Aimbient's ambient AI records patient arrival at and departure from the operating room as accurately as manual documentation across seven surgical service types.

File photo: Li Ka Shing Center for Learning and Knowledge, Stanford University School of Medicine
File photo: LPS.1, CC0 1.0 · source

Automatic operation timestamps as accurate as manual ones across seven surgical specialties, according to Aimbient

A Stanford Medicine-led study published in JAMA Surgery finds that Aimbient's ambient AI records patient arrival at and departure from the operating room as accurately as manual documentation across seven surgical service types. But the validation covers only these two timestamps at a single institution — and all findings are mediated through the company's own paid press release.

What Is Actually New

A study led by Stanford Medicine researchers and published in JAMA Surgery is the first large-scale, multispecialty validation of automated operating-room documentation, according to Aimbient's press release of November 18, 2026. The company Aimbient, formerly Surgical Safety Technologies, describes the study this way: the system matched human accuracy on key operating-room timestamps, with no statistically significant difference between automated and manual records across seven surgical service types (press release via Yahoo Finance, source 3d79816a).

It is worth underscoring where the evidence comes from: AIMag has not reviewed the research letter in JAMA Surgery itself. All figures and characterizations, including the company's formulation of "human-level accuracy," come from a paid press release and secondary coverage.

How the System Works

Over the eight-month study, Aimbient's system ran continuously in real time in 14 operating rooms and one procedure room at Stanford Health Care. The computer model automatically detects when patients enter and exit the operating room (source 3d79816a).

Accuracy was measured against blinded manual video review by humans — that is, people unaware of the system's records went through video footage and determined when the events actually occurred. The comparison was thus not whether the timestamps were "correct" in the abstract, but whether the machine's records matched what blinded human review judged to be the actual times.

Unlike previous validations of computer-vision-based operating-room timestamps, which according to the press release were limited to a single procedure type, this study included every surgery performed in the study rooms during the study period (source 3d79816a).

The Lag That Didn't Come Along

Perhaps the most concrete finding is not about accuracy in the absolute sense, but about time. Manually recorded timestamps lag the actual event by an average of five to six minutes, according to the press release. The reason is practical: the circulator nurse — the person who is supposed to enter the timestamps — is busy receiving and attending to the patient during exactly those minutes (source 3d79816a).

This means that "equal accuracy" is not the same as "equally timely." A manual timestamp can be just as correct when it is finally recorded, but a system that records in real time removes the built-in delay.

Why Timestamps Matter

Aimbient's founder and CEO, Teodor Grantcharov, explains the significance this way, as quoted by Unite.AI (Unite.AI, source 35635b5d): "Every OR metric a health system cares about, from first-case on-time starts to turnover time, is based on the accuracy of the timestamp underneath it."

In practice, this means that reports on operating-room utilization, wait times and efficiency — the numbers health executives use to allocate resources — rest on a timestamp basis that is currently recorded manually, with a built-in delay of several minutes.

Where the Evidence Ends

Several important limitations exist, and the Unite.AI reporting is explicit about them (source 35635b5d):

  • Only two milestones. The validation covers only the timestamps for patient entry and exit — not all of the surgical milestones in Aimbient's broader platform.
  • One center. The research letter describes the work as a single-institution experience across different procedures, and notes that studies of real-world implementation and validation remain scarce.
  • No effect data. Unite.AI explicitly distinguishes between what was measured (timestamp accuracy) and potential uses (efficiency gains, cost savings). Neither was measured in this study.
  • Already launched. Aimbient had already announced on July 10, 2025 that automated detection of surgical milestones and integration with electronic health records was in general availability (source 35635b5d). In other words, the product is in use ahead of broad validation.

Open Questions

Three questions remain: How well the system performs at hospitals other than Stanford Health Care. Whether the remaining milestones in the platform hold the same accuracy. And what the research letter's actual methods, statistics and author list show — which requires independent access to the JAMA Surgery publication itself, not just the company's description of it.

AIMag.no
AIMag.no
The AIMag.no editorial team covers artificial intelligence, tools, research, and regulation.

Get the best of AI MAG in your inbox

News, analysis, and ideas at the intersection of AI and society.