How is liability for clinical AI being defined and distributed?

Key takeaways

As more health systems put clinical AI into use, physician groups are pushing back on the idea that clinicians alone should bear liability when an AI-assisted decision contributes to harm. The American Medical Association wants liability to rest with the parties best positioned to manage the risks, including developers, while the FDA has expanded enforcement discretion for some clinical decision support tools. Sen. Bill Cassidy’s PATIENT AI Act discussion draft and a new Federation of State Medical Boards workgroup are both meant to produce recommendations on how to divide that accountability.

Clinical AI is officially entering its “not it” era.

After a relatively brief period of experimentation and vetting of AI-enabled clinical decision support and prediction tools, more and more health systems (more than a quarter, in one recent survey) are starting to put these capabilities into action.

These organizations are undoubtedly doing their due diligence to ensure these tools are as safe and effective as possible. But at the same time, they need to be asking themselves an equally important question. What happens when, not if, something goes wrong?

Fundamentally, it’s not a new issue for healthcare. Legal liability and ethical accountability have always been central to the practice of medicine, and there are well-established policies already in place to handle most situations.

However, AI introduces variables that existing frameworks haven’t accounted for. Can a machine be held accountable? Who pays the penalties when humans and computers make decisions together? How much responsibility can developers truly bear when their products are designed to adapt their outputs to what users ask of them? And how much falls on the individual clinician in charge of the patient?

These are the questions that policymakers are racing to answer, especially as adoption rates continue to outpace the implementation of strong, comprehensive governance at both the national and local levels.

Defining the role of the clinician in AI-enabled decision-making

Leading physician organizations have been very vocal defenders of the clinician as the ultimate authority on patient care decisions, even as they embrace AI as a valuable tool for improving the process.

Yet they are much less enthusiastic about the notion of clinicians bearing sole responsibility for the consequences when an AI-assisted decision contributes to a poor outcome.

For example, the American Medical Association’s (AMA) current policy, reiterated in a new issue brief, is that “liability for AI-related harm should rest with the parties best positioned to know the risks and to prevent or mitigate them.”

While the Association acknowledges that physicians are responsible for the “clinical judgement they can meaningfully exercise,” the AMA also firmly states that “developers of autonomous AI systems with clinical applications (screening, diagnosis, treatment) are in the best position to manage issues of liability arising directly from system failure or misdiagnosis and must accept this liability with measures such as maintaining appropriate medical liability insurance and in their agreements with users.”

Placing too much of the liability burden on individual clinicians could backfire on both the health systems adopting AI tools and the vendors trying to sell them, added AMA CEO John Whyte, MD, MPH, in a letter to the Washington State Medical Commission earlier this year.

“[A] physician who is told that they will bear full responsibility for a diagnostic miss produced by a flawed algorithm they did not know or have reason to know was experiencing a performance issue has a straightforward incentive: use AI tools as little as possible, or not at all,” he cautioned.

Where will developer accountability come from?

While physician-focused stakeholders may be pushing for more developer accountability, the federal government’s largely hands-off approach to AI oversight has the potential to complicate the issue.

The FDA has taken steps to pull back on regulatory oversight of certain clinical decision support (CDS) tools. In January 2026, updated guidance stated that the FDA would expand its policy of “enforcement discretion” for some CDS tools.

Some tools that allow clinicians to review recommendations before taking action are likely to be subject to reduced oversight, suggesting that the active exercise of clinical judgement plays an important part in accountability for outcomes.

If the FDA is not exercising its full authority to support developer oversight, what mechanisms will be available to the industry to ensure fair and equitable distribution of liability in this complicated environment?

Weighing a balanced approach to implementing accountability

Congress has the potential to fill in some of the gaps.

Starting in January, a newly Democrat-controlled legislature is likely to take a different position on AI oversight than the Republican-led Congress of the past two years.

But it’s a longtime Republican Senator, Dr. Bill Cassidy of Louisiana, who has already initiated the conversation on how to appropriately enforce accountability for the healthcare industry with a discussion draft released this autumn. Cassidy is leaving the Senate when the new Congress is seated, so the draft would need another sponsor to move forward.

If taken up by the chamber, the PATIENT AI Act would require HHS to start exploring how liability for AI-related medical errors should be divided among clinicians, healthcare organizations, insurers and technology developers.

The effort would culminate in recommendations for federal and state policymakers on potential changes to malpractice, product liability, licensure and insurance laws.

The Act would also clarify which technologies are subject to oversight by either the FDA or the Office of the National Coordinator for Health IT so that higher-risk tools receive the right level of scrutiny while offering users appropriate pathways for feedback and reporting.

“We’re living in a new digital age as artificial intelligence transforms how we deliver care. Implementing AI in health care must be done responsibly and to the benefit of patients, innovators, and consumers,” said Cassidy in a statement. “This discussion draft proposes solutions to expand access to AI-based technologies by creating regulatory predictability while prioritizing patient safety.”

The Federation of State Medical Boards (FSMB) is taking action, too, by convening the new Workgroup on the Regulation of Artificial Intelligence in the Practice of Medicine.

The group intends to develop recommendations or model guidelines for state medical boards about how to regulate AI tools used in clinical practice, including practitioner qualifications, oversight responsibilities, informed consent and standards of care for AI-enabled clinical practice. FSMB will also look at tools with “limited or no” direct physician supervision, which may include direct-to-consumer AI services and tools deployed within the EHR.

Moving forward into the next phase of AI accountability

The industry is likely to coalesce around some degree of shared accountability and liability, especially as the lines between the human and AI components of care continue to get blurrier.

In its October 2026 issue brief, the AMA said it would like to see liability frameworks include certain protections for human practitioners, including policies that shield clinicians who appropriately engage with AI, particularly when they couldn’t reasonably detect or prevent an error.

The Association would also welcome policies that prevent developers from shifting liability through disclaimers alone, as well as rules that protect patients’ ability to seek compensation from developers if they experience demonstrable harm.

Although the FSMB is still in the process of developing recommendations, previous policy documents highlight its position that physicians remain ultimately responsible for their clinical decisions, including when they choose to follow or reject AI-generated recommendations.

However, it also recognizes that developers and healthcare organizations are responsible for ensuring the safety, transparency and appropriate use of these tools, including providing clinicians with sufficient information, training and ongoing performance assessments.

Overall, the challenge will be finding the right blend of point-of-care accountability and upstream regulation of AI-enabled technologies to ensure that clinicians can retain meaningful authority over patient care decisions without being overburdened with liability for the actions of increasingly complex AI systems.

Jennifer Bresnick is a journalist and freelance content creator with a decade of experience in the health IT industry. Her work has focused on leveraging innovative technology tools to create value, improve health equity, and achieve the promises of the learning health system. She can be reached at [email protected].