AI Is Changing Surgery Fast, and Liability Rules Can’t Keep Up

A neurosurgeon writing in BW Healthcare World this month argues artificial intelligence (AI) in the operating room is amplification: extra data, sharper imaging, earlier warnings. Physician AI use jumped from 38% to 66% in a single year, American Medical Association survey data show.

Hospitals and insurers are moving slower than the surgeons already using these tools. Liability, consent, and who answers when a machine’s recommendation goes wrong remain mostly unsettled, even as adoption climbs across nearly every specialty.

What AI Actually Does Once the Surgeon Scrubs In

The surgeon’s argument turns on one distinction: pattern recognition happens in the data, while judgment happens in the exam room, with a patient’s family sitting across the desk. One surgeon framed the math simply: fifteen years of personal caseload is a small sample next to the tens of thousands of recorded procedures an algorithm can scan.

Researchers describe a similar split. AI can help identify critical anatomical structures, track instruments in real time, and flag deviations from established surgical technique, according to a recent review in a neurology journal. None of that requires the software to decide anything on its own. It requires the surgeon to see more information, faster, before making the same calls they always made.

UC San Diego Health became the first health system on the West Coast to perform AI-guided robotic spine surgery this spring, using a system that overlays a 3D reconstruction of the patient’s spine before implants go in. “AI-driven planning and patient-specific implants enable personalized surgical plans to enhance patient functional outcomes,” said Alexander Khalessi, chief innovation officer at UC San Diego Health and chair of its neurological surgery department.

• Before the incision – AI reconstructs a patient’s anatomy from scans, flagging tumor margins and vessel proximity a surgeon can study before touching an instrument.
• During the procedure – navigation software tracks instruments in real time and signals when they near a critical structure, pulling imaging and endoscopic views onto one screen instead of several.
• After the operation – predictive models trained on thousands of prior cases estimate a patient’s complication risk, shaping recovery plans before problems appear.

That structure, information first and decisions last, is why the surgeon calls the technology an assistant rather than a threat to the job itself.

The Adoption Curve

Physician adoption did not creep. It jumped.

Doximity, a professional network for U.S. physicians, tracked adoption climbing to 63% of physicians by early 2026, up from 47% a year earlier. Ninety four percent of the physicians it surveyed said they were either already using AI or wanted to be.

Health systems tell a similar story, and the specialty breakdown shows where interest concentrates hardest:

Surgery is not broken out as its own line in most of these surveys, but it sits inside every one of them. The da Vinci figure alone shows how far a single robotic platform can reach across two decades of use, according to the American College of Surgeons.

Does Pattern Recognition Count as Clinical Judgment?

Not automatically. AI can map a tumor’s exact margins, flag its proximity to the basilar artery, name its molecular subtype. Deciding whether to operate at all, how much risk a family should accept, what quality of life means for this particular patient: that still sits with the surgeon in the room, built through conversations that can run for hours.

The surgeon’s essay uses a brainstem glioma as the test case: an aggressive tumor, a poor statistical prognosis, and a decision that has nothing to do with whether the scan is accurate. The question is whether surgery is worth attempting at all, and what the family is prepared to navigate afterward. AI can inform that conversation. It cannot have it. The surgeon calls building that trust “the central act of what we do.”

The Patient Comfort Gap

Patients like AI’s presence right up until it starts making the call itself. That pattern shows up clearly in a 2026 scoping review of clinician and patient perceptions published in a peer-reviewed surgical journal.

• Patients report 72.9% comfort with AI issuing real-time alerts during surgery, but that falls to 47.5% for partially autonomous systems and just 17.7% once a system makes the call on its own.
• General surgery residents are wary of leaning on AI for structure identification during laparoscopic or robotic work, with only 12% in favor.
• Trauma surgeons split down the middle on whether AI genuinely improves surgical vision, at 53% agreement.

Some patients in the review described these systems as a “black box,” unsure how a machine reached its recommendation or whether it accounted for a case unlike any it had seen before.

Who Pays When the Algorithm Gets It Wrong?

Nobody has fully settled that question. Doctors, hospitals, device makers, and insurers each have a reason to want liability to land on someone else, and the first lawsuits tied to AI-assisted care are already moving through courts.

The American Medical Association has asked Congress to shield physicians from malpractice claims tied to AI recommendations. AMA President Jesse Ehrenfeld said the organization was “seeing lawsuits already,” according to a legal analysis tracking liability exposure for AI device makers. Manufacturers, in turn, face their own exposure once a device’s software plays a role in a bad outcome.

Banks are running into a version of the same problem. Financial crime compliance teams are discovering they need operating models built before AI deployment, not after, and hospitals are learning the identical lesson in real time.

States are not waiting for Washington. One 2026 policy tracker counted more than 250 AI-related healthcare bills introduced across at least 34 states in a single year. Colorado’s law, enforcement of which begins June 30, 2026, requires disclosure whenever AI shapes a major healthcare decision, plus annual bias audits and three years of recordkeeping. Utah has required upfront AI-use disclosure since 2025, with fines of $2,500 per violation. Texas requires plain-language disclosure whenever AI influences a high-risk healthcare decision.

Regulators Are Writing the Rules as Surgeons Race Ahead

The FDA now reviews these devices under a total product life cycle framework that tracks performance long after initial clearance, paired with predetermined change control plans that let companies push software updates without seeking new approval every time.

Robot-assisted surgery has climbed to what researchers call Level 4 autonomy, meaning high but not complete independence. Level 5, full autonomy, remains experimental, per the American College of Surgeons. In one case the college’s bulletin cited, a surgeon in London performed a robotic-assisted prostatectomy on a patient in Gibraltar, more than 1,000 miles away, with negligible delay.

A 2026 survey of healthcare and life-sciences organizations found generative AI overtaking data analytics as the top focus area, cited by 65% of respondents, with predictive analytics close behind at 51%. Data privacy and sovereignty topped the list of obstacles at 39%, followed by regulatory and ethical concerns at 37%.

The technology lowers the bar of who can do very complex procedures, so the learning curve is faster.

Bohdan Pomahac, chief of plastic and reconstructive surgery at Yale Medicine, said that of newer robotic platforms entering specialty surgery, in comments reported by MedTech Dive.

One live test of the surgeon’s broader thesis is already recruiting. A multicenter trial coordinated by China-Japan Friendship Hospital is set to begin enrolling stroke patients in April 2026, testing AI-guided robotic treatment of brainstem hemorrhage to see whether earlier, algorithm-assisted intervention changes outcomes the way predictive medicine promises.

Frequently Asked Questions

Will AI Replace Surgeons?

Most surveyed physicians don’t expect full replacement, though many expect the job to change. A Sermo poll found 58% of physicians believe AI will change the face of healthcare by narrowing the physician’s role or, in some views, making parts of the job obsolete. One neurosurgeon responding to that poll worried the bigger risk sits elsewhere: AI could let nurse practitioners and physician assistants be perceived as interchangeable with physicians, giving insurers a reason to replace doctors outright in certain settings.

How Accurate Is AI at Catching What Surgeons Might Miss?

Very, on narrow tasks. A 2025 review in a peer-reviewed surgical journal found AI tools reaching up to 97.5% accuracy in tumor detection and cutting resection errors by roughly 30% in the procedures studied. A tool called DeepGlioma classified tumor molecular subtypes in real time with more than 90% accuracy, fast enough to inform decisions mid-surgery rather than days later.

Who Is Legally Responsible When a Surgical AI System Causes Harm?

That is still being worked out, and it depends on where the failure happened. A white paper from the Society of American Gastrointestinal and Endoscopic Surgeons, published in Surgical Endoscopy, proposes a tripartite framework for surgical AI risk: flaws inherent to the AI system itself, mistakes introduced by the clinician using it, and failures that arise from how the hospital deployed it. Courts have not agreed how to divide liability across those three buckets.

What Is the Difference Between a Robotic Surgery System and an AI Surgery System?

Not every robot in the operating room runs AI, and not every AI tool involves a robot. Platforms like the da Vinci Xi already use non-AI robotic features such as tremor reduction and movement scaling that smooth a surgeon’s hand motion without any algorithm deciding anything. Fully autonomous surgical AI, where software acts without a surgeon’s hand on the controls, does not exist yet in approved practice; machine-vision features meant to flag unseen blood vessels or nerves in real time are still emerging.

How Does the FDA Actually Approve These Systems?

Most surgical robots reach the market through the FDA’s 510(k) pathway, which requires a company to show its device is substantially equivalent to one already approved, rather than proving safety and effectiveness from scratch. That pathway works for incremental updates. It gets harder once a system claims a genuinely new form of autonomy, since regulators then classify it under a stricter category requiring a full premarket approval application.

Disclaimer: This article is for informational purposes only and does not constitute medical or legal advice. AI-assisted surgical tools carry clinical and liability risks that vary by procedure, jurisdiction, and device. Patients and clinicians should consult qualified medical and legal professionals before making care or compliance decisions. Figures cited are accurate as of publication and may change as regulations and studies evolve.