AI Malpractice Risk Grows as FDA Clears 1,451 Devices

What's happening

By early 2026, the story of artificial intelligence and medical malpractice is not one of courtroom chaos but of quiet infrastructure-building. Thousands of AI tools are now embedded in American clinical practice, insurers are beginning to ask pointed questions about how those tools are governed, and lawyers are mapping out liability theories — but the lawsuits themselves, and the pricing data that would follow them, largely have not materialized yet 71415.

The clearest evidence of scale comes from the FDA's own tracking. Through the end of 2025, the agency had authorized 1,451 AI-enabled medical devices since it began keeping records in 1995, with 1,104 of them — 76% of the total — in radiology 6. In the fourth quarter of 2025 alone, the FDA cleared 72 AI-enabled devices, 55 of which were imaging products 6. GE HealthCare leads all companies with 120 radiology AI authorizations, followed by Siemens Healthineers, Philips, Canon, United Imaging, Aidoc and DeepHealth 6. That volume matters to malpractice insurers because imaging AI sits directly inside diagnostic decisions — the kind of missed cancer, hemorrhage or fracture that has always generated negligence claims 6.

Against that backdrop of rapid deployment, the malpractice claims record itself remains thin. Deepika Srivastava, chief operating officer at The Doctors Company, told Medical Economics that there has been "no documented malpractice case in the U.S. where AI was central to the claim," and that most insurers are seeing only limited AI-specific claims so far 7. Jared Kaplan, CEO of malpractice insurer Indigo, makes a similar point in blunter terms: the loss data needed to price AI risk precisely "doesn't exist yet," because malpractice claims from care delivered in 2025 and 2026 are years away from resolution, while clinical AI has only been in widespread use for about two years 15.

The liability landscape taking shape

Legal and clinical analyses converge on a similar map of where AI-related liability could eventually land. A physician might be exposed for accepting an AI output without independent review (automation bias), for ignoring an AI-generated alert, for overriding a correct AI recommendation, or for using a tool outside its cleared population or indication 9. Hospitals face exposure for failing to validate a tool locally, train staff, or monitor performance after deployment 79. Vendors, meanwhile, occupy a comparatively sheltered legal position for now, according to Medical Economics, even though a ScienceDirect narrative review and a University of Michigan law-school paper both argue that developers should eventually bear more responsibility when an algorithm operates with real autonomy 7817.

A University of Miami law review essay pushes this furthest, arguing that as AI-enabled devices move toward full autonomy — the "robotic surgeon" that operates without human intervention — Congress and the FDA should preemptively build a liability framework rather than let litigation sort it out case by case 16. That is a forward-looking, somewhat speculative argument; nothing in the other sources suggests regulators are close to adopting it.

A recurring theme across the legal sources is that FDA clearance is not a safe harbor. The physicianaihandbook.com compliance guide states this most explicitly: FDA authorization does not establish that the local standard of care was met, does not decide negligence or causation, and does not protect a physician from liability 9. The FDA's own device-classification pathways — 510(k) substantial-equivalence clearance, De Novo classification, and PMA approval for higher-risk devices — carry different legal consequences; the Michigan law review notes that PMA-approved devices are more likely to shield manufacturers from state tort claims under federal preemption doctrine, but only three AI-based devices have gone through PMA, since the vast majority are cleared through the less rigorous 510(k) route 17.

The regulatory groundwork

The FDA has been building toward this moment for years, from its 2019 discussion paper and 2021 AI/ML action plan through a series of guidance documents on good machine-learning practice, predetermined change-control plans, and transparency 10. Its January 2025 draft guidance on AI-enabled device lifecycle management — detailed by both the FDA itself and by law firm Dentons — asks manufacturers to document model design, data management, bias testing, performance validation and post-market monitoring, explicitly flagging that AI systems are vulnerable to "data drift" as real-world use diverges from training conditions 101112. Comments on that draft closed in April 2025, and the physicianaihandbook.com compliance guide notes the FDA finalized its predetermined-change-control-plan guidance in August 2025 and revised its CDS guidance in January 2026 — though the AI-device lifecycle document itself remained in draft status as of that account 911.

Where the reporting agrees

Across the trade press, insurer commentary, legal scholarship and the FDA's own materials, there is strong consensus on several points. First, no outlet claims that a documented AI-caused malpractice case has yet worked through the U.S. court system — Srivastava's statement to Medical Economics and Kaplan's remarks to getindigo.com both describe an absence of resolved claims 715. Second, every legal source agrees that FDA clearance does not equal a malpractice defense or a settled standard of care 891617. Third, insurers are described consistently as cautious rather than aggressive: riders and sublimits are appearing, particularly in Europe, but blanket AI exclusions are still rare, and The Doctors Company says it currently has none 714. Fourth, radiology's dominance of the FDA's AI clearance list — around three-quarters of authorizations in recent years — is corroborated by the imaging-industry tracking and cited as context by the broader liability literature 67.

Where it doesn't

The most consequential divergence is over how much weight AI actually carries in the current premium-hardening cycle. The AMA's 2026 policy analysis, drawing on Medical Liability Monitor data covering 60% to 80% of the market, documents that 39.9% of reported premiums increased in 2025, up sharply from 13.7% in 2018, with double-digit jumps concentrated in states like Pennsylvania, Rhode Island and Kansas 13. That data set makes no mention of AI as a driver at all — it reads as a story about tort environment, verdicts and inflation. Yet the headline framing of this topic implicitly invites readers to connect AI to malpractice-insurance costs. The tension is genuine: outlets covering AI adoption (the imaging trade press, legal reviews) treat the technology as a looming structural risk, while the one source with actual premium numbers shows a market whose recent hardening predates widespread clinical AI use and is not attributed to it anywhere in that document 613.

A second, subtler disagreement concerns directionality. Kaplan, whose company sells malpractice coverage, argues that AI is more likely to reduce claims than increase them, because many malpractice cases stem from fatigue and attention lapses that automation doesn't share — comparing it to Waymo's driverless-mile safety record 15. The legal scholarship is more agnostic or even leans the other way, framing AI as creating new failure modes (opaque decision-making, distributed responsibility, a "responsibility gap") that existing negligence law struggles to handle 81617. Kaplan has an obvious commercial incentive to frame AI favorably for underwriting purposes, which the getindigo.com pieces do not flag but readers should weigh.

A third point of difference, more a matter of emphasis than fact, is who bears near-term liability. Medical Economics states plainly that physicians and health systems are left carrying the bulk of liability while manufacturers operate in a sheltered position 7. The Michigan and Miami law reviews largely agree but treat this as an unstable, temporary arrangement that should shift as devices grow more autonomous — a normative argument rather than a description of current law 1617.

The read

The evidence does not support a claim that AI is currently driving medical-malpractice premiums higher — the AMA's granular data undercuts that narrative even as trade coverage of FDA clearances makes AI's clinical footprint look enormous. What the sources do jointly establish is a market in a holding pattern: regulators are building lifecycle-documentation requirements, insurers are collecting underwriting intelligence rather than repricing risk, and lawyers are pre-litigating theories for cases that haven't yet been filed in meaningful numbers. The most defensible synthesis is that AI is becoming an underwriting variable, not yet a premium driver — and whether it eventually pushes rates up or down will depend on claims data nobody currently has.