Managing Risk and Liability When Patients Rely on ChatGPT and Other AI Tools

Gerald Zarlengo, MD, Chairman & CEO, Copic Insurance Company
Chairman & CEO Copic Insurance Company

Generative artificial intelligence (AI) and large language model tools such as ChatGPT and Claude are rapidly becoming part of the clinical encounter. Patients increasingly arrive at appointments with AI generated summaries of symptoms, differential diagnoses, and recommendations for testing or treatment. 

From a risk management and liability perspective, these encounters warrant particular attention. They affect expectations of care, perceptions of attentiveness, shared decision making, and—critically—how disagreements over testing or treatment are understood and documented.

WHY AI-INFORMED PATIENTS CREATE NEW LIABILITY EXPOSURE

Recent clinical commentary highlights that patients often use AI tools not to replace physicians but to prepare for visits and to advocate for themselves—particularly after feeling dismissed or unheard in prior encounters. When those advocacy efforts are met with defensiveness or dismissal, the risk is not merely dissatisfaction, it is erosion of trust, which remains a leading contributor to malpractice claims.¹

Several liability-relevant dynamics are emerging:

• Explicit Requests for Testing or Treatment: AI tools frequently suggest diagnostic studies (e.g., tilt table testing, advanced imaging, lab panels) without weighing patient-specific probability, sequence, or resource constraints.1
• Perceived Authority of AI Output: The polished tone and medical fluency of large language models can lead patients to overestimate accuracy and underestimate uncertainty.²
• Heightened Sensitivity to Refusal: When clinicians decline AI-suggested interventions, patients may interpret the decision as disregarded rather than clinical judgment—especially if communication is rushed or overly technical.¹
• From a claims standpoint, these encounters increase exposure to allegations of failure to order tests, failure to diagnose, or lack of informed decision-making—even when care aligns with evidence-based guidelines.

COMMUNICATION AS PRIMARY RISK-REDUCTION TOOL

The literature underscores a central finding relevant to risk management: patients value being heard and recognized at least as much as they value clinical accuracy. In medicolegal terms, acknowledgment is not a courtesy—it is a protective behavior.1

RECOMMENDED OPENING FRAMEWORK

Before addressing the merits of AI-generated information, clinicians should explicitly acknowledge the patient’s effort and concern:

• “I can see you’ve put a lot of thought into this.”
• “Let’s walk through what you found and how it applies to your situation.”
• This approach lowers defensiveness and creates a record of shared engagement rather than unilateral decision making. 

INTEGRATING ACCURATE AI-GENERATED INFORMATION SAFELY

When AI-generated content aligns broadly with medical knowledge, it can be incorporated into the encounter in a way that supports clinical authority and reduces risk.

Best practices include:

1. Acknowledge partial accuracy.
   “That test is used in certain situations, and it’s reasonable to ask about it.”
2. Reframe around individualized risk assessment. Emphasize probability, timing, and clinical context rather than absolutes.
3. Clarify sequencing and thresholds. Many liability claims arise not from refusing tests outright but from failing to explain why now is not the right time.

Documenting this discussion demonstrates that the clinician considered the patient’s input and applied professional judgment—an important defense if decisions are later questioned.

DOCUMENTATION: A CRITICAL
LINE OF DEFENSE

From a liability standpoint, documentation should reflect:

• That AI-generated information was discussed
• That patient concerns were acknowledged
• The clinical rationale for accepting or declining recommendations
• Evidence of shared decision making or informed refusal when applicable

Avoid chart language that implies irritation or dismissal (e.g., “patient insists,” “patient demands”). Neutral phrasing such as “patient inquired about…” or “patient requested discussion of…” is preferable.

Patients increasingly arrive armed with information to be heard. Meeting them with recognition rather than resistance preserves both the human core of medicine and the legal safeguards that support it. ν

Eligible Colorado Medical Society (CMS) members can receive up to a 10% premium discount on Copic policies. Get more information on CMS membership by contacting membership@cms.org. 

Sundar KR. When Patients Arrive with Answers. JAMA. 2025;334(8):672–673. doi:10.1001/jama.2025.10678

https://www.medscape.com/viewarticle/chatgpt-your-clinic-whosexpert-now-2025a1000lqt?ecd=a2a

The information provided herein does not, and is not intended to constitute legal, medical, or other professional advice; instead, this information is for general informational purposes only. The specifics of each state’s laws and the specifics of each circumstance may impact its accuracy and applicability, therefore, the information should not be relied upon for medical, legal, or financial decisions and you should consult an appropriate professional for specific advice that pertains to your situation.