Whodunnit? Finding Who Is to Blame When the Algorithm Gets It Wrong
With artificial intelligence (AI) being developed at an exponential pace by the day, regulations and legislations are struggling to keep up with preventing and addressing the consequences. In the medical industry, AI is being used to analyze data and speed up experimentation, an efficient solution in hindsight, but unintended consequences raise questions on patient protection and the ultimate accountability.
By BRANDON KOO
Consider the hypothetical. A hospital’s AI tool reviews a head CT and marks it as normal. A busy radiologist agrees, and the patient is discharged. However, days later, the same patient returns with a serious bleeding from the brain resulting in lasting damage. Now, the patient’s family demands answers from the hospital as to what could have possibly happened that led to a severe misdiagnosis. Could it be the radiologist who signed the report based on AI’s feedback, the hospital that bought the AI, the company that built it, or no one at all?
Why the Culprit Is Hard to Find
Typically, responsibility for clinical AI harm can fall on the developer through product liability, on the institution through vicarious liability and/or corporate negligence, or even on the clinician through medical malpractice. The more actors involved in the decision chain, the harder it becomes to assign responsibility to any one of them. Additionally, one could expect AI misdiagnosis cases to involve parallel malpractice and product liability claims, with each side pointing at the other: physicians may argue the software was defective, while manufacturers could argue that the physicians relied on it unreasonably and with overdependence.
However, in reality, courts have little precedent to draw on. Currently, very few decisions exist and liability is highly unlikely to rest on a single party.
Suspect One: The Radiologist
Under current law, a physician is often the default suspect. The Royal Australasian College of Physicians warned that physicians face significant risk of carrying primary responsibility for patient harm under existing frameworks, while on the other hand, a learned intermediary doctrine would place a manufacturer’s duty to warn on the physician instead of the patient.
Mock-jury research hints at how real juries might view the case. A study in 2026 developed a related scenario involving a missed brain bleed that the AI had correctly flagged. Nearly 75% of mock jurors found that a radiologist who reviewed the scan only once after the AI check failed their duty of care. Jurors were also 50% more likely to side with the plaintiff in that case than when the radiologist read the scan twice, once before and once after seeing the AI output.
Suspect Two: The Hospital
Before the AI device is even used, the hospital chooses the device, designs the workflow and then signs the vendor contract. Oftentimes, indemnification clauses, liability caps and warranty disclaimers in those contracts will most likely shift risk back onto the healthcare organization. Due to this matter, employers and service providers such as hospitals could potentially be among those who share liability with developers, vendors, clinicians and users, through indemnity, contractual and insurance arrangements.
Suspect Three: The Manufacturing Company
Alternatively, responsibility could be tracked to the very developer with questions of whether the manufacturing company or developer should be held accountable for flaws in products that results in a risk of human life being unsettled. Though toys and food items are undoubtedly considered “products”, courts have traditionally not treated pure software as a “product” under product liability law, and this has yet to be resolved for healthcare AI. Some commentators suggest that AI that replaces the physician in decision-making should instead be judged by malpractice standards. Contrasting viewpoints assert that vendors’ “decision support only” label on the AI device can help them avoid FDA regulation and product liability exposure.
However, if AI becomes the standard of care, a catastrophic failure could expose a manufacturer to claims on a scale the industry has never seen. Thus, the stakes for developers, in other words, may rise as the technology becomes more central to care.
So, Who Is to Blame?
No direct answer can be determined at this point, and every allocation carries a cost. Placing most of the burden on clinicians may leave them accountable for devices they cannot fully inspect. Placing it on developers could discourage development of tools that help patients. And spreading the blame across the chain may leave a family struggling to learn who answers for what happened.
To prevent the hypothetical from becoming a reality, courts and legislatures must be able to settle whether AI counts as a product, how much scrutiny a physician owes an AI output and how shared responsibility should work.
Now, the question becomes a matter of if legal reform can adapt to the pace of AI development and how regulations should be implemented at the different levels of industries.
Sources & Further Reading
Niehoff, Lena. “New Product Liability Risks for AI Products.” Taylor Wessing, March 6, 2025.
Ramos, Frank. “Frank Ramos Explores Intersection of AI-Enabled Medical Devices With Liability.” Goldberg Segalla, April 20, 2026.
“Shifting Liability: AI in Medical Devices.” Crowell & Moring, February 19, 2020.
Shryock, Todd. “Can Physicians Be Sued for AI Mistakes?” Medical Economics, September 21, 2026.
The Royal Australasian College of Physicians. “Doctors Left Exposed as AI Outpaces Legal Protections, as Peak Medical Body Warns of Legal Black Hole.” The Royal Australasian College of Physicians, 2026.
Yu, Christine. “How AI Is Integrated Into Clinical Workflow Lowers Medical Liability Perception.” Penn State News, March 10, 2026.
Comments