Inside OpenVet: Five questions with our Chief Veterinary Officer, Dr. Natalie Marks
Dr. Natalie Marks serves as Chief Veterinary Officer of OpenVet. She is responsible for the clinical standards the platform is held to, including the boundaries on what our system will and will not put in front of a clinician.
She also leads the OpenVet Clinical Advisory Board, a founding panel of board-certified specialists in emergency and critical care, oncology, neurology, internal medicine, pharmacology, and general practice. The board provides direct clinical oversight of how OpenVet reasons across more than 200 species. As Dr. Marks has described the standard, clinical AI must be species-aware, evidence-based, and managed by veterinarians who deeply understand the life-and-death realities of the exam room.
Before joining OpenVet she spent two decades in clinical leadership, most recently as co-owner and medical director of Blum Animal Hospital, the largest small animal practice in Chicago.
We put five questions to her on the future of clinical AI in veterinary medicine. Her answers appear here in full, lightly edited.
Question 1: What do you believe about the future of veterinary medicine that most of your colleagues haven't come around to yet?
I believe veterinary medicine is about to stop being a memory-based profession. For my entire career, we have treated recall as the measure of a good clinician. Who can hold the most drug doses, the most differentials, and the most edge cases in their head at two in the morning? That is the model I trained under, and it is the one most of my colleagues are still quietly defending.
Here is what I think they have not fully accepted yet. Within a few years, the value a veterinarian brings will not be what they can recall. It will be the quality of the questions they ask and the judgment they apply to the answers. Recall is becoming a solved problem. Clinical intelligence systems will carry the literature, dosing, breed risk profiles, and conflicting guidelines, and they will carry them with citations attached. What they cannot carry is the relationship with the animal in front of you, the read on the client across the table, and the wisdom to know when the textbook answer is the wrong answer for this patient.
That is not a downgrade for the veterinarian. It is a promotion. We get to focus our cognitive energy on the part of medicine that is truly hard and truly human. But it asks us to let go of an identity many of us have held tightly for twenty years, which is genuinely uncomfortable. I think the colleagues who lean into that shift early will define the next era of this profession. Those who treat it as a threat will spend those same years on defense.
Question 2: There's no regulatory body that evaluates AI tools for veterinary use. What does that mean for practitioners trying to choose the right tools?
Right now, the veterinarian is the regulator. There is no FDA clearance to point to, no validation requirement, and no minimum standard for transparency or accuracy. Any company can build a veterinary AI tool, market it to clinicians, and put it in front of patients without anyone independently verifying that it works or is safe. The entire burden of evaluation has quietly landed on the individual practitioner, often without the practitioner realizing it.
That is an enormous and frankly unfair responsibility to place on a clinician who is already stretched thin. But until formal oversight arrives (and it will), it is the reality we are working in. So my advice is to evaluate these tools the way a regulator would, by asking hard questions before you trust any output. I tell veterinarians to ask five things of any AI vendor. Can I see the specific sources behind every recommendation, not a vague gesture toward the literature? Does the system tell me when it is uncertain or when my patient does not match its training population? Do I understand what data it uses as input and what it is missing? Has it been validated, and can I see the actual results rather than a marketing claim? And is it built specifically for veterinary medicine, or is it a general human tool with a veterinary skin on top?
I want to be clear that the absence of regulation is not only a danger. It is also an invitation. Companies that choose to build to a high standard now, publish their validation, and welcome scrutiny instead of avoiding it will set the bar that regulators will eventually adopt. Practitioners have more power here than they think. Every time a veterinarian refuses to trust a black box, they help define what trustworthy looks like for the whole profession.
Question 3: What would a truly trustworthy AI clinical tool need to do, and what should it refuse to do?
A trustworthy clinical tool must do one thing above all else. It must let the veterinarian independently review the basis for every recommendation. That is the line the FDA drew for human medicine in its decision support guidance, and I think it is exactly right for us, too. If the clinician cannot see why the system is recommending something, cannot access the evidence, and cannot judge whether it applies to their specific patient, then it is not decision support. It is a black box, and a black box has no place in clinical care.
In practice, a trustworthy tool cites specific texts, guidelines, and peer-reviewed work that I can verify, not a vague reference to veterinary literature. It tells me how confident it is and distinguishes between established facts and its own reasoning. It flags when the evidence is thin or conflicting. It tells me when my patient does not resemble the population it learned from. And it weights a published textbook more heavily than a random web page, because not all sources deserve equal trust.
What it should refuse to do is just as important. It should refuse to answer when a critical input is missing or ambiguous. The clearest example is species. Acetaminophen is a reasonable choice for a dog and lethal for a cat. A trustworthy system stops and confirms the species before it offers a dose, rather than guessing. It should refuse to present its reasoning as settled fact. It should refuse to fabricate a citation it does not have. Above all, it should refuse to make the decision. The moment a tool tries to replace the veterinarian's judgment rather than inform it, it has crossed a line it should not cross. The best clinical AI is confident about what it knows, honest about what it does not, and humble about whose call it ultimately is.
Question 4: Corporate veterinary groups are investing heavily in AI. Independent practitioners are using whatever's free. Where does that leave the profession in five years?
This is the question that keeps me up at night because it could go two very different ways, and the difference is a choice we are making right now.
The path I worry about is a two-tier profession. The corporate groups build or buy validated, integrated, well-governed AI, and their clinicians practice with a real safety net beneath them. Meanwhile, the independent veterinarian, the solo practitioner, and the rural mixed practice reach for whatever is free. And what is free today is general-purpose tools like ChatGPT, which are fast and often accurate but also opaque, uncited, and prone to inventing references. I do not say that as a criticism of the veterinarians using them. They are drowning in complexity, and nothing better has been within reach. But if that gap hardens, we end up with a profession where the quality of care an animal receives depends on the capital behind the clinic. That should trouble all of us.
The path I am betting on, and the reason I do the work I do, is the opposite. The same technology that could divide this profession is the most democratizing force we have ever had access to. Done right, the solo practitioner at two in the morning can have the same evidence-based clinical intelligence at her side as the largest corporate hospital in the country. Access to elite medicine does not have to track the size of the balance sheet. But that only happens if someone deliberately builds trustworthy tools for the independent practitioner and refuses to make transparency a premium feature. Five years from now, we will either look back on this as the moment the profession split or the moment it leveled. I know which one I am working toward.
Question 5: There's a tension between AI that delivers information and AI that acts as a clinical partner. Where should that line be?
The line is judgment. AI should expand the veterinarian's thinking, not replace it.
A good clinical partner does more than hand you a fact. It surfaces the differential you were too tired to consider, reminds you of the breed risk you half-remembered, and shows you the conflicting study you had not seen. That is partnership, and it makes us better doctors. But the instant a system moves from informing your decision to making it for you, it has crossed from partner to something else, something that needs a much higher bar and far more oversight than anything we have today.
The cleanest way I know to locate that line is the test of independent review. If the veterinarian can see the basis for a recommendation, weigh it against the patient in front of them, and choose to agree or disagree, the AI is a partner. The clinician is still flying the plane. The moment the system asks you to trust an output you cannot interrogate, the line has been crossed, no matter how helpful it feels in the moment.
I hold this firmly because of what is truly at stake in our exam rooms. The relationship among a veterinarian, a patient who cannot speak, and a family who loves that patient is close to sacred. AI should serve that relationship. It should never insert itself in the middle. The best partner is the one who makes the human in the room more capable, more present, and more trusted, and then gets out of the way.
Dr. Marks leads the OpenVet Clinical Advisory Board and oversees the clinical standards behind the platform. This conversation is one in an ongoing series from inside OpenVet.
OpenVet is free for licensed veterinarians.