‘An unstable situation’: Richard Anderson, M.D., FACP, on AI, liability and access through 2026

The system around American medicine is shifting faster than many physicians can track. Artificial intelligence (AI) is moving from the margins into everyday care, jury awards in malpractice cases are climbing into “nuclear” — and sometimes “thermonuclear” — territory, and consolidation is changing where and how patients can get seen. At the same time, the legal and regulatory frameworks that are supposed to govern all of this are struggling to keep up.

To make sense of that landscape, Medical Economics sat down with Richard Anderson, M.D., FACP, CEO of The Doctors Company and TDC Group, to talk about his organization’s annual report: “Healthcare on the Horizon: Predictions for U.S. Healthcare Through 2026.”

Below, Anderson discusses what he sees as the most underappreciated risks facing physicians today, why AI could both help and complicate clinical decision-making, how nuclear verdicts ripple far beyond individual cases and why access to care is already in a “rolling” crisis.

The following transcript was edited for style and clarity.

You open the report by noting that long-term forecasting is getting harder as the pace of change accelerates. From your vantage point, what is the most underappreciated risk physicians are facing in 2026?

I think this is a very insightful question. The pace of change in the universe is accelerating, but if we talk about the medical universe, every aspect of it is changing dramatically. Consolidation is changing the way medicine is practiced. Artificial intelligence has potentially very large inputs and very large potential for changing the way medicine is practiced.

It is 100% certain that the technology that is integral to the practice of medicine today — and certainly integral to digital medicine, but really all medicine, including AI — will not be matched by the legal system. The legal system will be completely out of step with the realities of current medical technology and the way medicine is practiced. The pace of change may be increasing everywhere, but not in the legal system.

It will take years of multiple separate trials on repetitive subjects to establish precedents that are widely recognized and that physicians and patients can understand as defining what the law is. All of this relates to two fundamental things. One is that a lot of the problems in medicine, particularly in the United States, are based on how we finance medical care.

The government now pays for more than 55% of all medical care in the United States, but that 55% actually does not pay for itself. It has to be subsidized by private payers, so private payers are paying much more for care. At the same time, after years of decline to record lows, the number of Americans who have medical insurance — something like 92% — is now starting to reverse. We will see more and more Americans without health care insurance.

Finally, all of this leads to burnout, and burnout is a serious problem. Nobody wants to see doctors who are frustrated, distracted, unhappy and not able to pay attention, with competing priorities. We are doing next to nothing about burnout. We talk a lot about it, but we tend to respond with nice things — the “kumbaya” things — like pizza on Friday or a monthly meeting to talk about issues for half an hour. That is fine, but the cause of burnout is basically the industrialization of medicine. It is about throughput.

As long as we prioritize throughput over quality of care and the doctor-patient relationship, burnout is not going to get better. There is promise in AI that it can help with some of these things, and perhaps we will talk more about AI. There is progress in the range of disease and illness that can be effectively treated by modern medicine, and some of that progress is very dramatic. But the system by which health care is practiced is, on one hand, being revolutionized, and on the other hand, being torn apart. That is a very unstable situation.

The report says AI will permeate health care, but that its value depends on clinician trust. Where does that trust most often break down, and why?

That is a very important question, and it goes right to the heart of what I mentioned already — the disconnect between our legal system and the way health care is practiced. There are more than 1,000 AI tools that have already received FDA validation. On the other hand, there are many more AI tools than those 1,000 that are being used in practice in a variety of different ways.

Physicians have essentially no ability to thoughtfully purchase or select among the thousands of AI applications — which ones are better, which ones are dangerous and which are not. The AI conundrum, dealing with thousands of these things, is a microcosm of one of the problems of American medicine, which is that there is very little actual standardization of care. We talk about “the standard of care,” but that is very different than saying care is standardized.

Every hospital, every doctor, has to make almost individual decisions about how AI is going to be used, and they have very little basis for doing it. Large systems have more ability to create dedicated teams to evaluate AI, but even there it is very difficult. Even if you evaluate an AI tool and it looks great, it is very difficult to evaluate how it fits into your ecosystem of care, because there are so many potential interactions.

The problem for physicians in terms of trust is the direct disconnect between the standard of care and AI recommendations. If AI makes a recommendation that is different from what the physician perceives as the standard of care, but it produces a better outcome or a good outcome, then there is really no problem because there is no damage and no lawsuit. Everyone is happy.

But if AI produces a recommendation that is different from the standard of care, the doctor follows it and the outcome is adverse, then by definition the doctor has violated the standard of care. Clinical adoption of AI is actually going to be slowed by this paradox. Using it for recommendations that are different from your clinical judgment exposes you to liability, whereas when you use your clinical judgment and follow the standard of care, our law says that if there is an adverse outcome, you are not liable. All things are not perfectly treatable, and an adverse outcome is not automatically a matter of liability.

So this is a real conundrum. Eventually, AI will become the standard of care, but how that happens will be piecemeal. It will be different venue to venue, and system to system. When we talk about AI, we all say “AI” as if we know what it means, but we are actually talking about an almost infinite system of recommendations, thoughts and data. It is not a single thing that will finally get perfected.

AI is continuously learning and morphing — hopefully for the better, but not necessarily. Of the 1,000 tools already approved by the FDA, is there any way to monitor what they are doing now and how they are being used? No. All of this creates a very uncertain situation. I am absolutely certain that incorporating AI will become the standard of care, and when AI becomes the standard of care, however we get there, that paradox will be resolved. Follow the recommendations, and you are practicing according to the standard.

A lot of physicians feel trapped: rely on AI and risk liability, or ignore it and risk being seen as negligent. Practically speaking, how should physicians be using AI today to protect patients and themselves?

It is a very good question to which there really is no simple answer. The most widely used, or at least the most widely cited, example of AI is ambient listening, in which an AI agent monitors the doctor-patient interaction, records the voice exchanges, organizes them into a coherent accounting of the encounter and enters it into the medical record.

Most of the groups and individuals using the more sophisticated ambient listening programs are quite pleased with them, because they effectively solve the problem that developed with the widespread adoption of the electronic health record. For all its promise, physicians were spending several hours a day, often after work, trying to enter the required data into the record.

Ambient listening allows most of that to be done by an automated system, and the doctor simply has to affirm that the note is accurate. On the other hand, it is a little unnerving that the solution to a technological problem created by electronic health records is another, more sophisticated technology. We have been down that road before. It would be nice if every new technology did not require another technology to monitor and regulate it. We will see.

The report describes a “trillion-dollar migration” to digital-first care. What kinds of digital investments most often create unexpected legal or clinical risk once they are deployed?

You can look at telemedicine as a kind of parent. In telemedicine there is a human being; the digital version of it is basically telemedicine without a human being. If you are going all the way in that direction, you have a kind of “AI telemedicine.”

We are going to have to distinguish who is responsible. If you have an “AI physician,” what does that even mean? What are the limits of that AI physician, and how is that AI physician licensed? If you are a medical practitioner and you practice medicine without a license, that is criminal. If you are a computer practicing medicine without a license, I assume it is illegal, but it seems to be happening.

How is it licensed? How is it regulated? If the computer messes up, who are we suing? Presumably we will end up, as our legal system does, blaming the doctor the AI is substituting for. Somewhere in that loop is a human physician, and that physician will be sued.

It is much more challenging to sue Google or Anthropic or any of the smaller AI companies proliferating everywhere. Some large plaintiffs’ firms will want to sue them, because they presumably have big dollars behind them. On the other hand, those cases will drag on in court for many years. That means a long time before resolution, a long time before precedents are established, and when precedents are established, they are only in one venue and subject to appeal.

There have been dramatic advances in medicine over the last 125 years, going back to the scientific basis of contemporary medicine. With digital innovation, it is difficult to know what that progress looks like. It is difficult to measure it, to know what outcomes are acceptable and what are not, and to integrate human medicine with the way medicine has been historically practiced.

Again, there is a disconnect between the rapid changes affecting medicine today and the court system, which knows how to sue physicians, hospitals and established institutions, but has very little precedent for suing “agentic” things. One example is that there is not much litigation against electronic health records, even though they have been a mixed blessing at best. The litigation is based on physicians and institutions that use them.

The report draws attention to “nuclear verdicts” — plaintiff awards exceeding $10 million. They’re becoming more common and more extreme. What has changed about juries, legal strategies or society that physicians may not fully appreciate?

Nuclear verdicts are a very serious problem. In just the last two or three years, if you take the average of the 50 largest medical malpractice judgments in the United States, that number has increased from a little over $30 million to more than $56 million per case. The outlier judgments beyond that — the so-called “thermonuclear” judgments — now regularly exceed $100 million each year.

Why is that happening? There are many reasons. One is society. The distribution of wealth has created a kind of monetary desensitization. It used to be that $100,000 seemed like a lot of money. Then millions seemed like a lot, then billions. Now we are comfortable talking about trillions, even though virtually none of us can comprehend what a trillion anything is.

Younger juries have different perceptions of what is fair and equitable. They place less faith in institutions — whether government, physicians, attorneys or retail — because they have grown up in a world of constant change and constant legal attacks. All of that affects how juries approach these problems.

There is also a cultural problem that is not generational: the notion, very strong in the United States, that any adverse event that befalls us is someone’s fault and that person should be held accountable and compensate us. If you do not sue, you are “leaving money on the table.” That is a very dangerous attitude.

Ultimately, we end up suing ourselves. As it is now, 55% of all health care in the United States is paid for by the government, and there is no reason to believe that percentage will not continue to go up. As the government pays more and more of health care, it becomes literally true that we are suing ourselves. If the government pays the indemnity, that means taxpayers pay it — and we are the taxpayers.

Each nuclear verdict is huge and has a large impact on an insurance company or institution, but the effect goes far beyond the individual case. A $50 million or $100 million verdict resets the range of value that society assigns to similar cases. If there is a $50 million verdict for an anesthetic death of a child in one venue, the range for discussion in similar cases in that venue will be at least $50 million.

Most claims are settled for less than that, but the expectations are reset. It anchors plaintiffs’ and plaintiffs’ attorneys’ expectations of what a case is worth, even if the facts are different. So nuclear verdicts are a problem in and of themselves, because they represent hundreds of millions or billions of dollars in a year, but the effect, like a wave, goes throughout the entire health care system.

How close are we to an acute access crisis, and what does that mean for physicians practicing in those communities?

In a sense, we are in the middle of a rolling access crisis. For years we have had concerns about the number of Americans who lack health insurance. When you lack health insurance, you either lack access to care, or when you get care and pay for it out of pocket, you can easily end up in bankruptcy. The most common cause of bankruptcy in most parts of the United States is medical bills.

As medicine consolidates and large corporations take over many medical institutions, including hospitals, they may decide it is not financially viable to operate a hospital in a rural area. Or, if they do, they will only provide services that are profitable or have limited liability.

About 60% of rural hospitals in the United States no longer provide obstetrical care. That is an extraordinary gap. A roughly similar percentage of rural hospitals do not provide at least one other form of specialized care. The safety net in rural areas is becoming dangerously frayed, and it is part of a larger access-to-care issue we have not solved.

The report says tort reform could become a national priority. If physicians and medical societies could realistically move the needle on just one or two reforms, what would you recommend?

We now have 50 years of experience with legal reforms in medical malpractice, going back to MICRA — the Medical Injury Compensation Reform Act in California in 1975, which is really the granddaddy of all medical malpractice tort reforms. It has established a model for the country.

The one thing that is abundantly clear is that the most effective single tort reform is a limit on non-economic damages — so-called pain and suffering. The reasons are multiple. Pain and suffering is very difficult to quantify. What is grief worth? What is pain worth?

We want a system that is relatively fair to everyone. How can we have a system in which a similar injury to a similar individual is worth $50 million or $75 million in one venue and $2 million in another? It is inherently inequitable, yet that is what our system is.

More fundamentally, in states that have no caps on non-economic damages, we are effectively insuring infinity. You cannot collect enough money from physicians and hospitals to accommodate verdicts of $500 million or $900 million. There was a recent $900 million medical malpractice verdict. How do you insure that? You really cannot.

Insurance is integral to American life and to the way health care functions. We need to protect the viability of the insurance system. Reasonable tort reform is necessary so that one patient and one plaintiff’s attorney do not take a disproportionate amount of money out of the health care system. Health care is a zero-sum game. If we take $100 million for one case, that is $100 million not available for the system as a whole.

When an AI system acts on its own and something goes wrong, who do you expect courts to hold responsible?

For sure, somewhere in the loop there is a doctor or a hospital, and I guarantee you they are going to be involved in the suit. It will be much more challenging, until precedents are worked out, to figure out the responsibility of the AI.

Is the AI really practicing independently? How can the AI practice independently if it does not have a license? How can it have a doctor-patient relationship? What is the training that certifies it is qualified to do what it says it is doing, versus a physician? What is the standard of care for an AI?

People often use autonomous driving as a metaphor. When an autonomous vehicle has an accident, it is front-page news, even if autonomous vehicles may be safer than regular drivers. Every time an autonomous vehicle crashes, it is a story. That is similar to AI in medicine.

Eventually AI will be part of the standard of care, but how that takes place when we have hundreds or thousands of AI systems and hundreds of thousands of care venues is unclear. There is no seamless integration and no certification of AI that fits into our health care system. We mentioned 1,000 AI products approved — that is different from saying they will not have adverse outcomes.

Physicians are left with a paradox. If they follow AI recommendations and there is no adverse outcome, everyone is happy. If they choose not to follow an AI recommendation and the outcome is adverse, there is an argument they will be sued for not using AI. There is no good answer until society comes to grips with how our legal system works, what we are willing to delegate to machines and where we insist on human judgment.

We talk about having a “human in the loop.” To the extent that we have humans in the loop, I guarantee you the humans will be sued. We still have not figured out how we are going to be suing AIs and what that litigation will look like.