Surgery Insight Report
Surgeon perspectives on where intraoperative AI is useful today, what happens when software conflicts with tactile judgment, which real-world conditions expose model limits, and why trust also depends on workflow, training, and accountability.
Audience: Surgeons
Countries: 6
Survey records: 88
-Hero findings
– Quick Read — Key Findings
The full report follows how surgeons verify conflict, interpret failure, assign accountability, and define the safeguards needed for AI in the OR.
Computer vision, automated phase recognition, image overlays, and AI-assisted analytics are moving closer to live surgical workflows. That creates a different trust problem from retrospective decision support. An intraoperative prompt appears while the surgeon is acting, often in a field shaped by distorted anatomy, inflammation, scar tissue, bleeding, smoke, multiple instruments, and limited time to reconsider the next step.
In June 2026, AORN released an evidence-based guideline on integrating artificial intelligence into surgical care. The guidance addresses patient safety, governance, bias, privacy and security, education, and clinical oversight, while reinforcing that AI is intended to support clinical judgment rather than replace it.
Clinical context: Association of periOperative Registered Nurses (AORN), Guideline for Integration of Artificial Intelligence, June 2026.
MDForLives surveyed surgeons across the United States, United Kingdom, Canada, Italy, France, and Germany. The pattern is transitional rather than binary: many operating suites remain at pilot stage or have no integration, yet surgeons already have clear views on where AI adds utility, what makes it unreliable, how they respond to disagreement, and what governance they want around it.
MDForLives Research Interpretation
Only 8.8% report advanced, natively integrated real-time computer vision or AI overlays. Another 17.6% use assistive or external analytics, while 27.9% are at pilot stage and 45.6% report no AI integration.
What the pattern suggests: The profession is being asked to form trust, liability, and training expectations while many surgeons still have limited exposure to deeply integrated intraoperative AI.
What may be behind it: Adoption depends on hardware, software integration, institutional governance, reimbursement, and data handling, not only on whether an algorithm performs well in validation.
Why it matters: The debate is no longer purely futuristic, but neither is it fully routine. That gap can make governance decisions especially consequential because policies may be forming before broad workflow experience accumulates.
Surgical phase recognition leads at 43.3%, ahead of anatomical boundary mapping at 25.4%, automated post-operative benchmarking at 19.4%, and intraoperative crisis early warning at 11.9%.
What the pattern suggests: The strongest perceived value sits in workflow recognition rather than AI taking over a high-stakes surgical decision.
What may be behind it: Phase recognition can support logistics and situational awareness without requiring the surgeon to surrender control of tissue handling or accept an autonomous intervention.
Why it matters: The first durable role for intraoperative AI may be narrower than the technology narrative. Tools that organize, label, or quantify the case may earn trust before tools that tell the surgeon where or how to cut.
40.3% pause and seek a secondary maneuver or alternative imaging. Another 19.4% consult a colleague or assistant before overriding the warning. 29.9% proceed on manual intuition, while 10.4% change trajectory purely for defensive medicine and liability mitigation.
What the pattern suggests: Disagreement becomes a trigger for more information rather than automatic deference to either the machine or the surgeon’s first impression.
What may be behind it: Intraoperative decisions are irreversible in a way many diagnostic decisions are not. Verification provides a way to preserve surgeon authority while acknowledging that an algorithmic warning may contain useful information.
Why it matters: The clinical value of an AI alert may depend as much on the escalation pathway around disagreement as on the alert itself.
54.5% identify severe inflammation, scar tissue, atypical anatomy, or other distorted anatomical conditions as the most frequent cause of AI diagnostic failure. Instrument tracking interference follows at 24.2%.
What the pattern suggests: The algorithm is most vulnerable when anatomy itself departs from the clean or expected pattern, not simply when the camera view is imperfect.
What may be behind it: Severe inflammation, scarring, prior surgery, or atypical anatomy can change landmarks and tissue planes in ways that challenge pattern recognition, precisely when surgical complexity is already high.
Why it matters: A model that performs well in routine anatomy may still need clear uncertainty behavior in the cases where the surgeon most wants assistance.
36.5% expect primary liability to remain with the attending surgeon after an overridden AI warning and a complication. 31.7% expect responsibility to become blurred, 22.2% place it on the manufacturer, and 9.5% on the hospital credentialing committee.
What the pattern suggests: Surgeons do not see the presence of AI as removing human accountability. At the same time, nearly one in three expect liability to become harder to assign.
What may be behind it: A logged warning creates a new record of disagreement. That can make the decision to override feel different from a conventional judgment call, even when the software is advisory.
Why it matters: Liability ambiguity can create defensive behavior, slow adoption, or encourage over-response to alerts unless organizations define how AI outputs are documented and reviewed.
Shared-liability frameworks lead at 38.7%, followed closely by mandatory “unplugged” training at 35.5%. Dynamic artifact filtering and instant override controls receive smaller shares.
What the pattern suggests: The leading safeguards sit around professional accountability and preservation of manual skill, not around a single technical fix.
What may be behind it: Automation bias is also framed as a training problem. 43.5% identify over-reliance by younger residents as the greatest risk, while 29.0% point to alert fatigue among experienced surgeons.
Why it matters: Trust in intraoperative AI appears to require a system around the model: credentialing, training, override authority, legal clarity, and workflow design.
Many surgeons are still outside deep AI integration, yet the survey does not show blanket resistance. Instead, surgeons distinguish between operationally useful tools, higher-stakes real-time guidance, and the safeguards required around each.
When software and tactile judgment disagree, the dominant response is to add evidence or another human perspective. The hidden pattern is procedural trust: confidence depends on what happens after the alert, not only on the alert’s accuracy.
Distorted anatomy leads the reported failure conditions. That shifts attention from average model performance toward robustness in unusual, scarred, inflamed, or anatomically complex cases.
Liability, manual skill development, workflow integration, reimbursement, and data ownership all sit beside technical performance. In this survey, AI adoption is as much an institutional operating model as a software decision.
The survey does not point toward a simple replacement story. 35.5% expect routine, highly repetitive components of surgery to become autonomous while complex reconstruction remains manual, 25.8% foresee the surgeon becoming an intraoperative supervisor of automated steps, and 29.0% expect the role to remain largely unchanged with AI restricted to a secondary safety function.
The most consistent signal is therefore not a vote for or against AI. It is a preference for bounded automation. Surgeons appear most comfortable when the system has a defined task, uncertainty triggers verification, the surgeon can override it, and accountability is clear before a complication tests the policy.
This matters because the barriers are also practical. Integration with native video or EHR workflows and reimbursement each lead the adoption barriers at 30.6%, while data sovereignty and privacy follow at 25.8%. The route from an impressive algorithm to a trusted intraoperative tool runs through workflow and governance as much as through model performance.
Direct answers to the questions healthcare professionals are most likely to ask about these findings.
Clinical context sources: AORN Guideline for Integration of Artificial Intelligence. These FAQs provide general educational context and do not replace specialty guidance or patient-specific clinical judgment.
