Ambient listening systems, predictive analytic algorithms, automated documentation — these are just a few examples of how artificial intelligence (AI) is being used in healthcare, including physical therapy. The rapid expansion of AI usage has outpaced traditional legal and regulatory frameworks, creating uncertainty regarding professional liability.
AI has the potential to benefit both clinicians and patients by enhancing workflow efficiency, supporting clinical decision making, and improving outcomes, but AI tools also carry risks. Physical therapists (PTs) can safeguard themselves from liability by understanding these risks and taking steps to mitigate them.
AI liability risks
Mello and Guha, professors and experts who specialize in the intersection of law, technology, and healthy policy, have identified three scenarios for AI-related liability: (1) defective software used to manage care or resources, (2) harms that occur after clinicians rely on software for care decisions, and (3) malfunctioning software embedded in medical devices. The authors note that there is limited case law related to patient injury caused by AI. Part of the problem is the difficulty in assigning responsibility when harm occurs (the manufacturer, the clinician, the organization that decided how it would be used), making it challenging for plaintiffs to file and win claims. The “black box” nature of AI also makes it hard to demonstrate that an alternative design would have prevented the injury.
Of course, PTs and organizations want to avoid litigation even when it is not likely to be successful. In addition, clinicians remain accountable for delivering high-quality care. For example, if a PT based their treatment on inaccurate results from a cardiac screening without confirming the information with the patient, they would still be held liable if they neglected to consider the patient’s true cardiac condition, resulting in harm. Standards of care still apply.
Therefore, understanding AI risks is essential. An example of an individual PT risk is unauthorized recording and data capture using AI tools, such as what can occur with ambient listening software. This can lead to violations of privacy laws such as the Health Insurance Portability and Accountability Act (HIPAA). Failure to obtain consent to use AI also violates PTs’ ethical principles. Standard 2.2 of the American Physical Therapy Association Code of Ethics states that “Physical therapists and physical therapist assistants shall obtain ongoing informed consent after providing information that is understandable, honest, and necessary to allow the patient or client or their surrogate to make informed decisions about participation in physical therapist services or research.”
Perhaps more importantly, failing to obtain consent erodes patients’ trust when they discover that AI is being used without their knowledge. When patients lose trust, they may be more inclined to file claims, and they may resist treatment plans, hindering their progress.
Another individual risk is if PTs make faulty practice decisions when a predictive algorithm fails to identify a patient at risk for a certain type of therapy. This situation can lead to charges of professional negligence if the patient suffers harm.
At the organizational level, there may be potential liability issues related to malfunctioning AI tools that have been purchased, particularly if the organization has not addressed potential legal exposure in its contract with the AI vendor.
Reducing risk
PTs can take several steps to reduce their risk of liability related to AI.
Obtain consent for the use of AI. Chau and colleagues note that informed consent requires patients to receive clear, specific information about how AI will be used in their diagnosis and treatment, including benefits, risks, limitations, and data usage. Providing that information can be challenging, given the often-opaque nature of AI systems and clinicians’ limited familiarity with the AI details. Chau and colleagues provide a suggested template for an AI consent form that presents the information in plain language.
Patients may be most concerned about data sharing. Any data shared with other sources (such as the AI vendor to help improve the tool’s functionality) must be de-identified.
The signed consent form should be filed in the patient’s health record. In addition, PTs should know the process to follow if a patient declines AI consent.
Integrate AI with patient assessment. Do not rely solely on AI algorithms when treating patients, as each patient has unique needs. In addition, AI algorithms can have inherent biases because of the under-representation of data from certain demographics.
Verify data. Always verify information generated by AI systems. For instance, ambient listening systems might miss critical details or input incorrect information. Automatically generated notes could lack essential information or contain details that do not apply to the patient. Comprehensive and accurate documentation is the primary defense against a legal claim. If available, include the AI version number used for clinical decision making.
Check state laws and regulations. Laws and regulations related to AI usage vary by state.
Seek relevant education. Learn more about how to use AI effectively while minimizing liability risks.
PTs must remember that humans, not AI, are ultimately responsible for patient care. In addition to clinicians, organizations play a role in reducing AI liability risks (sidebar).
Risk vs. benefit
AI has the potential to improve patient outcomes and reduce PTs’ workload through enhanced documentation and information that supports their clinical decision-making. But AI also comes with liability risks, such as failure to obtain informed consent for its use and data misinterpretation. Fortunately, PTs can take steps to minimize their risks, including obtaining consent, integrating AI with patient assessment, and verifying data.
Lynn Pierce, FNP-C RN, Risk Management Consultant, HPSO
This article does not constitute legal advice.
Sidebar
Assessing organizational AI liability risk
AI tools currently require less testing compared with drugs and devices. Any problem with an AI model can affect many patients, creating the potential for significant liability. Therefore, organizations need to evaluate risks when considering an AI tool purchase. Mello and Guha identified four factors to consider:
- The likelihood and nature of errors. This is based on the AI model, its training data, its task design, and how it is integrated into the clinical workflow.
- The likelihood that errors will be detected by humans or another system before they harm patients. This partly depends on the amount of time and visibility humans have with the AI tool.
- The potential harm if errors are not caught. This is especially crucial for tools involved in critical clinical functions or used in the care of patients with serious health conditions.
- The likelihood that injuries would result in compensation through the tort system. This is influenced by factors such as the severity of the injury, ease of proving negligence, and the causal relationship between the AI tool and the injury.
Organizations need to carefully review contracts with AI vendors before signing. For instance, vendors might include indemnification clauses that shift responsibility for errors to the organization. Consider using a vendor vetting checklist: https://cms.laptboard.org/assets/Newsroom/Integrating-AI-Tool-into-Your-Practice-Checklist.pdf.
In addition, organizations should conduct audits of AI performance after implementation.
Sources: Mello M. Michelle Mello: Understanding liability risk from healthcare AI tools. Stanford HAI video. https://www.youtube.com/watch?v=P7Z1yV-PAjQ&t=3779s; Mello MM, Guha N. Understanding liability risk from healthcare AI. Stanford University Human-Centered Artificial Intelligence. Policy Brief: HAI Policy & Society. 2024. https://hai.stanford.edu/policy/policy-brief-understanding-liability-risk-healthcare-ai; State of Louisiana Physical Therapy Board. AI in physical therapy: ethical use, legal risk, and patient trust. 2025. https://www.laptboard.org/news/ai-in-physical-therapy-ethical-use-legal-risks-and-patient-trust
References