Have you ever used a tool at work that you didn’t fully understand, but used anyway because everyone else was? That’s one of the biggest threats with AI in medicine right now.
For a while, the conversation about AI and patient safety was mostly hypothetical. What might go wrong someday. What we should probably watch for eventually.
That changed in March 2026.
ECRI, a nonprofit patient safety organization, released its annual report with the Institute for Safe Medication Practices. They ranked the ten biggest threats to safe care in the U.S. this year.
Above rural hospital closures. Above the return of vaccine-preventable diseases. Above federal funding cuts.
Number one was diagnostic AI.
Not future AI. Not some theoretical version down the road. The tools being used in clinics and hospitals right now, today, on real patients.
Disclaimer: While these are general suggestions, it’s important to conduct thorough research and due diligence when selecting AI tools. We do not endorse or promote any specific AI tools mentioned here. This article is for educational and informational purposes only. It is not intended to provide legal, financial, or clinical advice. Always comply with HIPAA and institutional policies. For any decisions that impact patient care or finances, consult a qualified professional.
What the problem is
The report isn’t saying these tools are bad. That’s not the argument.
The real issue is the gap between how fast we’re using them and how well we actually understand them. A survey of nearly 1,200 physicians found that 66% reported using AI in clinical practice in 2024. That’s up from 38% the year before. Almost double, in twelve months.
The tools spread faster than the training did. Faster than the policies. Faster than anyone figured out who’s accountable when something goes wrong.
And the performance numbers are where it gets real. Tested machine learning models failed to recognize 66% of critical or deteriorating conditions in synthesized cases. Popular generative AI tools saw their accuracy drop when the prompts came from open-ended patient conversations instead of clean, textbook descriptions.
Sit with that one for a second.
A tool that performs well on a tidy, structured input but struggles with the messy, incomplete, sometimes contradictory way patients actually talk is a tool worth understanding closely before leaning on it. Performing well on a test and performing well in the room aren’t always the same thing.
ECRI itself frames the risk in three parts: more diagnostic errors, more bias, and an erosion of our own critical thinking over time. Three different problems. All three are real.
Who takes the liability?
Richard Anderson, CEO of The Doctors Company, put it plainly to Medical Economics:
“If AI makes a recommendation that’s different than the standard of care, and the doctor follows it, and the outcome is actually adverse, then, by definition, the doctor has violated the standard of care.”
Read that again.
That’s the paradox Anderson is describing: the tool makes the call, but under this reasoning, the liability can still land on the physician who followed it. Worth knowing, whatever your specialty.
Anderson said more than 1,000 AI tools have already gotten FDA validation, but most of us have no real way to evaluate which ones are actually reliable, or what happens when one gets it wrong. His words, not mine: “It’s 100% certain that the technology that is integral to the practice of medicine today, which includes AI, the legal system will not keep up with that technology.”
So that gap isn’t closing anytime soon. Which is part of why ECRI’s guidance points toward understanding a tool’s limitations at the individual level, not just assuming it’s been sorted out somewhere upstream, whether it came from your hospital, your EHR, or a vendor’s pitch deck.
What’s being recommended
A few things, and they may apply to you individually, not just to the hospital system.
First, clear AI usage policies are lagging behind adoption. The AMA’s 2026 survey found that 81% of physicians now use AI professionally (more than double the 2023 rate) while physicians consistently rank data privacy assurances and validated safety/efficacy as the top prerequisites for trusting a tool. In practice, that gap between fast adoption and slower institutional guardrails means you’re often the one evaluating a tool’s reliability yourself, without a formal policy to lean on.
Second, training on the specific tools you’re actually using. Not AI in general. Knowing a tool has FDA clearance tells you almost nothing about where it degrades or which patients fall outside what it was validated for.
Third, documentation. Write down when AI influenced a clinical decision. ECRI frames this as good practice on two fronts: it creates a record, and over time it gives your practice real data on whether a tool is actually performing the way it was advertised.
Fourth, and this one matters most: treat AI as a supplement, not a replacement. That’s not a policy line. That’s just a habit you build.
Unlock the Full Power of ChatGPT With This Copy-and-Paste Prompt Formula!
Download the Complete ChatGPT Cheat Sheet! Your go-to guide to writing better, faster prompts in seconds. Whether you’re crafting emails, social posts, or presentations, just follow the formula to get results instantly.
Save time. Get clarity. Create smarter.
Here’s the part I don’t think gets said enough
It would be easy to walk away from all this thinking AI is just risky and we should all proceed with our arms crossed.
That’s not the full picture. The same report that ranked diagnostic AI as the top safety concern also said, clearly, that AI has real potential to improve how we work and who gets access to good care. Both things are true at once. A tool can be genuinely useful and still carry risk that deserves real attention.
I think the physicians who come out ahead here are the ones willing to hold both of those truths. Leaning on AI without understanding its limits carries real risk. But so does avoiding it entirely just because the oversight isn’t perfect yet, since colleagues who learn to use these tools well may end up with real efficiency gains.
ECRI put this at the top of a ten-item list this year, which suggests it reflects a pattern their team is seeing show up across the health care system, not a one-off concern.
Taking it seriously doesn’t mean walking away from AI. It means applying the same rigor you’d apply to any other clinical tool. Know what it’s good at. Know where it breaks. Write it down when it influences a call. Keep your own judgment in the loop, always.
That’s not some new, higher bar we have to clear.
It’s just medicine. Same as it’s always been, applied to a new kind of tool.
So, how are you handling AI in your own practice right now? Are you using it, avoiding it, or somewhere in between? Let us know in the comments.
At Passive Income MD, we cover the tools, strategies, and practical AI workflow tips helping physicians build more time and financial freedom. We’ll keep tracking where AI goes from here.
Download The Physician’s Starter Guide to AI – a free, easy-to-digest resource that walks you through smart ways to integrate tools like ChatGPT into your professional and personal life. Whether you’re AI-curious or already experimenting, this guide will save you time, stress, and maybe even a little sanity.
Want more tips to sharpen your AI skills? Subscribe to our newsletter for exclusive insights and practical advice. You’ll also get access to our free AI resource page, packed with AI tools and tutorials to help you have more in life outside of medicine. Let’s make life easier, one prompt at a time. Make it happen!
