Understanding what an AI tool can and cannot do is only part of the challenge endocrinologists face. Consideration must also be given to how its use affects documentation, privacy, regulatory requirements, professional liability, and responsibility for decisions made at the point of care. 

In the final part of his conversation with AACE Endocrine AI Editor-in-Chief Johnson Thomas, MD, FSCE, FEAA, Steven Petak, MD, JD, MACE, FACP, turns his attention to practical guidance for incorporating AI into patient care. He discusses common pitfalls physicians should recognize and steps for using AI tools responsibly.

(Note: The information provided in this interview is for general informational and educational purposes only. It is not intended to constitute and should not be relied upon as legal advice. The following transcript has been edited for clarity and length.) 

Dr. Thomas: Let's talk a little bit about practical advice. Can you give us two or three habits an endocrinologist should build today, or start doing, to stay on the right side of AI and medicolegal practice? 

Dr. Petak: First, you have to know about the tool. You need to know its name and version, and you need to document whether it's FDA authorized. You need to understand the patient population in which the tool was validated and, critically, its major limitations. You don't need to understand every line of code, but you need to know what the clinical promise is that you're relying on. You need to document your judgment in the follow-up. You need to determine why this fits or doesn't fit this particular patient, and why you made that decision. And this is a particularly important area, such as incidental vertebral fractures that are being found, thyroid molecular results, and automated insulin delivery systems in particular. 

Thirdly, you want to discuss privacy issues. You want to tell the patient when AI may materially affect their care when the law requires it, because states differ. Do not enter protected health information into an AI service unless the organization has approved its use and has the proper business associate agreement in place. When you're using some of these public AI tools, please do not put any identifying information in there. And if the patient has a rare disease, that may be enough to really identify that patient in that region that you are practicing in.  You have to be really careful, unless you're using AI tools that are HIPAA compliant and have a compliant business associate agreement. 

Dr. Thomas: How does the practicing endocrinologist keep up with changing and different state and national laws? Are our societies helping us understand these laws? 

Dr. Petak: That's a great question. Busy endocrinologists can't become 50 state legislative services. Healthcare systems need to centralize that function, so you really need to know your jurisdiction. You need to know the laws in that jurisdiction, but the use of state medical societies, specialty organizations, counsel if you have an organization that has inside legal counsel, and policy reviews are really important. This is not something most individual positions can do. It's really an organizational responsibility, not a scavenger hunt for each position. These are issues that really need to be centralized in bigger systems and not be burdened with the individual position. 

Dr. Thomas: We all carry malpractice coverage and policies, individually or through our institution. Are you seeing malpractice policies excluding these AI-related medical devices or medical services that are related to AI? Are you seeing these in malpractice policies? 

Dr. Petak: They are, but there's not going to be generally a paragraph that says, "artificial intelligence." You're going to see paragraph language that says things like "definitions of professional services," "exclusions for experimental or unapproved technologies," "cyber and privacy issues," and "third-party software or notices." This is all going to be policy-specific, and it's all over the map right now. I think it's really important to clarify your coverage before an incident occurs. This is better than trying to litigate what the definition of professional services is afterward. Don't assume that AI is covered or excluded. You want to try to get that answer in writing before a potential claim occurs. 

Dr. Thomas: Most physicians that I know are well-intentioned, and they practice medicine to the best of their ability. What's the single most common mistake you'd think a well-intentioned physician is making right now with AI? 

Dr. Petak: The most common mistake is really confusing what looks like a fluent output with a reliable one. AI can produce confidence faster than it can produce truth. It's critically important to note that signing a note, for example, without confirming that it actually accurately reflects the encounter is a huge issue. Copying an AI-generated summary into the chart without verifying it, accepting recommendations without checking whether it fits that patient, overlooking results that conflict with the clinical picture, or placing identifiable patient information into an unapproved service are really critically important. You want to pause and then ask, what's the evidence? What could make this wrong for this patient, and what am I going to do next? You want to verify that the final record is really complete before signing it. AI can certainly accelerate reasoning and documentation, but it cannot replace verification. 

Dr. Thomas: Five years out, do you think that the wave of AI lawsuits in endocrinology will land with the endocrinologist, or the device manufacturers, or is it going to be a mix? 

Dr. Petak: It'll land on everybody. Manufacturers and vendors have the deepest pockets, so lawyers are definitely going to want to keep them central. Insurers, though, will face claims when algorithms wind up supplanting physician judgment. And healthcare systems themselves will be questioned about things like validation, training, workflow, and monitoring. But I do expect physician-focused claims to grow as AI becomes more embedded in your daily care. The allegation is not going to be, "you created a defective algorithm," it's going to be something like "you selected the wrong patient, ignored a warning, accepted a recommendation without clinical correlation, or failed to disclose something that was material and didn't follow up." I don't think the algorithm itself is going to be the issue; it's going to be how we employ it, and the follow-up care and the safety issues that are going to be there. AI is not going to choose one defendant; they're going to follow who's controlling the system. 

Dr. Thomas: What are some of the developments that you're most excited about and closely following? 

Dr. Petak: It's really happening right now, where we're moving from decision support to autonomous AI. Our legal system is really built around the assumption that the physician makes a decision, and you're using technology as a tool. And that assumption falls apart when software independently does things, like screen, diagnose, communicate with patients, adjust therapy, or initiate actions. The key questions are really who's in control, who has a realistic opportunity to catch an error, and whether law and regulation will impose direct duties on all of these players. It's going to affect everything. It's going to affect informed consent, FDA oversight, malpractice standards, insurance coverage, what it means to be a human in the loop. It's where the technology and the legal doctrine are most likely to collide. And the basic question is, it's no longer whether AI gives advice. The question is really whether or not AI is acting. And I think that's where things are going to change, and they are changing right now. 

Dr. Thomas: I'm going to ask a cliché question. Do you think AI will replace us in the future? 

Dr. Petak: Endocrinology is uniquely suited to AI, because our field is rich in data, images, risk models, and longitudinal decisions, which are very nicely handled by AI systems. AI is going to help detect these diseases earlier, personalized treatment, these patterns in glucose adjustments and bone—in particular finding vertebral fractures—are clearly going to help with patient management. The physician's role isn't going to go away, but it's going to shift. I think we'll be doing less manual calculation and more interpretation, deciding whether the data are reliable, whether the model fits this particular patient, how the results should change care, and communicating of uncertainty and choices. It's not that AI will replace us, the danger is really that we'll become passive and less thoughtful. 

Instead of really thinking through problems, we'll be relying on AI to the point that we will no longer be on top of our own care of the patient. You need to really use proper AI well. You want to use AI care to make it proactive and personal, but used poorly, it can make error rates faster and harder to see. I think it'll move endocrinologists from calculators to curators, interpreters, and decision makers. And I think we won't be replaced, but our roles are going to change. 

Dr. Thomas: What's the one thing that you want the endocrinologist or a physician to do differently in how they use AI for patient care, so they can start implementing that in their practice? 

Dr. Petak: You want to engage with AI, but you do not want to defer to it. The next time you have an AI tool that affects patient care or diagnosis, don't simply accept it or reject it. Engage with it and make that engagement visible. Ask whether the tool was built for this particular clinical scenario and this particular patient. Compare it with your own history, examination, and laboratory data. Then write one clear sentence in the notes stating the tool and version used, why you agreed or disagreed with it, and the follow-up plan. That single habit really captures a single principle. AI may contribute a lot of the evidence, but the physician still owns the clinical judgment and the continuity of care. AI is our partner, but don't let it take over your judgment.