Authored by: Kendall Barrett
Photo Credit: Catherine Caruso, An AI System with Detailed Diagnostic Reasoning Makes Its Case, Harvard Medical School (Oct. 8, 2025), https://hms.harvard.edu/news/ai-system-detailed-diagnostic-reasoning-makes-its-case.
The use of artificial intelligence (“AI”) in the practice of medicine is no longer a speculated future but the current reality. Today, this technology is embedded in the operational reality of healthcare in the United States. As these AI technologies become routine tools in patient care, they raise important and unresolved questions that courts and legislators are just beginning to grapple with. How does the emergence of AI in healthcare affect the standard of care? Who can be liable when an AI-generated response contributes to a patient’s harm? How should courts evaluate AI evidence?
AI integration in medicine is moving at a rapid pace. Even as early as 2020, AI assistance in clinical settings spanned areas such as diagnostics, imaging analysis, triage optimization, and predictive risk analysis.[1] This growth has since accelerated significantly, and a 2025 report from the JAMA Summit on Artificial Intelligence catalogued AI tools across nearly every medical subspecialty.[2] However, the legal system lags behind these developments. Medical malpractice doctrine centered around individual professional negligence, healthcare provider judgment, and the standard of care, reflecting a world where humans made all material clinical decisions and therefore could be held accountable for them. Now, AI models can initiate or guide these clinical decisions, obscuring the traditional notion of medical malpractice liability where these models are involved.
A healthcare provider who falls below the applicable standard of care and thereby causes harm to a patient may be liable. The standard of care has traditionally been defined by reference to what a reasonably competent practitioner in the same specialty would do under similar circumstances, but precise definitions vary by state.[3] AI involvement in treatment only further complicates this determination. Concerns about liability under the current medical malpractice doctrine incentivizes healthcare providers to follow the standard of care they originally would have, using AI as a confirmation tool only, rather than relying solely on the AI suggestions.[4] As AI technology develops and improves, its use in healthcare will likely become even more accepted.[5] Some suggest that following AI suggestions in treatment may become part of the standard of care in the future, so providers who refuse to utilize it may be liable for resulting injuries.[6] However, the current ‘two schools of thought’ doctrine allows for legitimate differences in treatment decisions if the different approach is recognized and supported by a respectable group of the medical community.[7] When AI is used in clinical care that leads to an adverse outcome, the initial question is whether liability belongs to the provider who used the tool, the health care system that procured it, or the software developer who created it.[8] Under current doctrine, because the law holds physicians responsible for their clinical judgments and courts have generally not permitted physicians to avoid liability by attributing error to the tools or guidelines they relied on, the physicians who follow erroneous AI recommendations are exposed to potential liability.[9] However, while physicians using AI still face exposure to medical malpractice claims, the hospital systems that procure AI tools may also be exposed to institutional liability under negligent selection/retention or negligent supervision theories if they fail to govern them properly.[10] Regarding the software developers, many clinical AI tools are regulated as medical devices under the Food and Drug Administration and therefore require regulatory approval.[11] Some approved devices are protected by federal preemption, which can limit certain state-law product liability claims.[12] However, failure to comply with the regulatory requirements may expose these software developers to liability. Additionally, because of the intangibility of software, courts have been hesitant to apply product liability doctrines in claims against developers.[13]
Another important concern is the evidentiary questions surrounding AI’s complex characteristics. How does the Daubert framework apply to AI tools or evidence created by them? How will jurors react to the technical complexities surrounding AI? Unlike traditional scientific tools or methods, AI systems often update and learn over time, always changing. AI systems are structurally opaque, meaning that there is a lack of clarity that could explain how an AI tool generated a certain output from a certain input.[14] The methodologies are often inexplicable, even to the developers themselves, which is not particularly helpful for answering the questions of reliability or validity.[15] Turning to jury perception, jurors may have a “technology effect” bias based on the preconceived assumption or optimism that computer systems are inherently reliable.[16] On the other hand, jurors may feel skepticism or hostility toward AI in general. Effective advocacy in this context will involve challenges similar to those that arise in complex medical issues. In both situations, the subject matter is highly technical and unfamiliar, requiring attorneys to be capable of translating these complicated concepts into clear, comprehensible explanations. Increased attorney familiarity with AI technology will likely benefit this goal. Computer science expert witnesses who can explain how an AI system was trained, what data was used, and how it validated the output will also play important roles in these cases.
Understanding the concerns regarding medical liability and AI is crucial for the improvement and responsible use of these rapidly developing tools. As courts confront more cases involving AI in the medical malpractice context, clearer legal standards will emerge. In the meantime, attorneys and healthcare providers should remain attentive to these developments and engage in the discussions around the increasing role of AI in healthcare.
[1] Meryl Kornfield, The Health 202: Artificial Intelligence Use is Growing in the U.S. Health-Care System, Wash. Post (Feb. 24, 2020), https://www.washingtonpost.com/news/powerpost/paloma/the-health-202/2020/02/24/the-health-202-artificial-intelligence-use-is-growing-in-the-u-s-health-care-system/5e52f13188e0fa632ba81ec7/.
[2] Derek C. Angus et al., AI, Health, and Health Care Today and Tomorrow: The JAMA Summit Report on Artificial Intelligence, 334 JAMA 1650-64 (2025), https://jamanetwork.com/journals/jama/fullarticle/2840175.
[3] Nicholson W. Price II, Sara Gerke, & Glenn Cohen, Liability for Use of Artificial Intelligence in Medicine, in Research Handbook on Health, AI and the Law 150, 152 (2024).
[4] Id. at 153.
[5] Id.
[6] Id.
[7] Id.
[8] Id. at 151.
[9] George Maliha et al., Artificial Intelligence and Liability in Medicine: Balancing Safety and Innovation, 99 Milbank Quarterly 629-47 (2021).
[10] Price II et al., supra note 3, at 150, 157-58.
[11] Id. at 162.
[12] Id.
[13] Michelle M. Mello & Neel Guha, Understanding Liability Risk from Using Health Care Artificial Intelligence Tools, 390 NEJM 271 (2024).
[14] Maliha, supra note 9, at 629-47.
[15] Brian Mackenzie, Part Three: AI on Trial – Admissibility of AI-Generated Evidence. Justice Speakers Institute (Oct. 14, 2025), https://justicespeakersinstitute.com/ai-generated-evidence-admissibility-on-trial/.
[16] Willem H. Gravett, Judicial Decision-Making in the Age of Artificial Intelligence, 58 Law, Governance and Technology Series 281, 288 (2023), https://doi.org/10.1007/978-3-031-41264-6_15.