When patients suffer harm from artificial intelligence tools in healthcare, they typically have nowhere to turn except court. A new analysis of 31 legal cases in the United States reveals how AI systems deployed across medical facilities, insurers, and health departments have negatively affected patient care, and why current liability structures leave patients without meaningful recourse.
Researchers Gennie Mansi and Mark Riedl examined patterns across documented incidents and lawsuits to understand where and how AI failures occur in medicine. Their findings, published in Nature Health, show that patient harm stems not from physicians alone, but from a complex web of stakeholders. Hospitals, insurance companies, state agencies, and technology developers all shape how AI tools affect the people who depend on them. Yet legal accountability frameworks treat physicians as the primary gatekeepers, leaving other actors largely insulated from responsibility.
This misalignment creates a fundamental problem. Patients harmed by AI decisions in medical settings struggle to identify who should be held accountable or what legal pathway exists to challenge those decisions. The current system offers patients few practical options except litigation, and even litigation faces obstacles when the technology itself is difficult for lawyers and courts to examine.
Liability Structures and Patient Access to Justice
The core issue centers on how liability is allocated in healthcare AI systems. Medical law has historically placed responsibility on physicians to prevent and mitigate harms from new tools. But when AI systems make decisions about patient care, from diagnostic recommendations to coverage denials, many parties influence the outcome. Insurance algorithms that deny treatment, hospital software that flags safety concerns, and state health department systems that screen cases all affect what patients receive.
Current legal frameworks do not reflect this reality. They place the burden on individual doctors to catch AI errors, understand opaque algorithms, and prevent downstream harms. This leaves patients with limited avenues. Mansi and Riedl’s research suggests that liability structures need to expand to recognize the role of developers, insurers, facility operators, and government agencies in shaping how AI tools impact care.
Without clearer accountability paths, patients injured by these systems face a double barrier: proving harm and identifying the responsible party. Courts and legal counsel lack the technical expertise to scrutinize algorithm behavior. Patients may not understand which stakeholder made the harmful decision or possess the resources to pursue claims.
A Patient-Centered Approach to AI Accountability
Rather than centering physician responsibility, Mansi and Riedl propose shifting focus to patient-centered accountability. This means designing liability structures that acknowledge the multiple stakeholders influencing healthcare delivery, then holding each accountable for their specific role.
The researchers also identify a second pathway: designing AI tools themselves to enable contestation. If patients and their lawyers could more easily examine how an AI system reached a particular decision, they could build stronger legal challenges. Tools that provide clear explanations of algorithmic reasoning could serve as evidence in court and help advocates support patients seeking recourse.
This dual strategy, restructuring liability and redesigning AI tools for transparency, requires collaboration between lawyers and technologists. Neither community currently focuses on the patient’s ability to challenge decisions or obtain accountability. Lawyers rarely input their expertise into AI system design. Technologists seldom consider how their systems will perform under legal scrutiny or what information patients will need to contest harmful outcomes.
Litigation as an Imperfect Solution
The fact that patients have pursued legal action in 31 documented cases underscores the absence of other remedies. These cases span harm from diagnostic AI, coverage-denial algorithms, child welfare screening tools, and triage systems. Yet litigation remains an expensive, lengthy, uncertain process, one that most harmed patients cannot afford to pursue.
Mansi and Riedl’s work demonstrates that the medical, legal, and technology communities must recognize where AI harms occur and redesign accountability accordingly. Until patients have meaningful pathways to challenge decisions, contest AI recommendations, and hold responsible parties accountable, litigation will remain the only available recourse for those harmed by healthcare algorithms.
The implications extend beyond individual cases. How healthcare AI systems are designed, deployed, and governed will determine whether patients can meaningfully access justice when these tools cause injury.
