The following is a guest article by Michael Arevalo, Psy.D. Director of Clinical Strategy at Core Solutions
Picture a behavioral health clinician early in a career, reviewing an intake form on an ordinary morning. The presenting complaint reads clean: anxiety, disrupted sleep, some family conflict. Nothing unusual, on the surface. Then one phrase stands out, an offhand line from the person. “I’m not sleeping much, but that’s normal for me.” This might be nothing. Then again, it might be the start of something worth asking about further.
That instinct, the willingness to sit with discomfort rather than paper over it, is exactly what most AI systems struggle to replicate. I see this gap constantly in conversations about AI and behavioral health.
Why Agreeable isn’t the Same as Accurate
Most large language models are trained to be helpful, and helpful, in practice, usually means agreeable. Ask one a question, and it hands back something confident-sounding, with no room left for doubt. That’s a minor annoyance in most fields, but in behavioral health, it’s a real problem. A clinician’s job, especially early in a career, often comes down to catching the detail that doesn’t fit: a phrase in an intake note, a shift in someone’s affect a template would never pick up on. An AI tool that hands you a clean summary isn’t saving you time so much as it’s quietly deciding what didn’t matter.
This is where ethically built AI earns its name. It is a specific standard and one with real requirements: a system that stays transparent about how it reaches conclusions and keeps a named person accountable for what it produces while treating safety and privacy as basic requirements. In behavioral health, this can help to protect the therapeutic relationship itself and keep the clinician as the one who makes decisions.
Clinical judgment, treatment authority, and responsibility for outcomes stay with licensed professionals. An AI system can surface a pattern, flag a risk indicator, or draft a summary for review. What it cannot do is make the final call. The moment a tool renders a judgment a clinician cannot trace back, examine, or override, something important has been lost.
Where the Law Already Agrees
The law is starting to catch up. Illinois’ WOPR Act prohibits licensed providers from letting AI make independent clinical decisions. Nevada’s AB 406 goes further, barring AI from providing professional behavioral healthcare outright. Utah’s HB 452 requires mental health chatbots to disclose their use and limits how they handle sensitive data.
The FDA draws a similar boundary between support and decision-making in the form of a four-part test that determines whether clinical decision support software stays exempt from medical device regulation. The software cannot analyze signals or images on its own, and it has to work from existing medical information. It must support a clinical decision rather than direct it, and a clinician must be able to independently review how it arrived at any recommendation. Fail any one of those tests, and the software becomes a regulated device. Put simply: If a clinician cannot see how a tool reached its conclusion, that tool now answers to a regulator.
What This Means for AI Adoption in Behavioral Health
AI can give clinicians real-time feedback, and few fields need that relief more than behavioral health. An AI-driven EHR activates every capability starting at implementation. What matters most is how well an organization governs it.
Daily workflows are where that governance gets tested. Staff need dedicated time to review AI-generated insights before those insights shape care. Training should teach staff to notice when an output feels a little too clean rather than just how to operate the tool. The real test is whether clinicians still trust their own read when something in the record doesn’t add up.
Going back to that intake form example from the beginning: The clinician reviewing it did not need an AI system to tell them the chart looked fine. They needed one that would not have smoothed over the line that gave them pause in the first place. That’s the standard behavioral health must live up to.
About Michael Arevalo
Michael Arevalo, Psy.D., PMP, is the Director of Clinical Strategy at Core Solutions, a company that has spent more than 25 years developing behavioral health and IDD EHR technology and today supports care for more than 500,000 individuals. He brings a background in clinical psychology and project management to his work guiding how AI and other emerging technologies are evaluated and adopted across Core’s platform. Dr. Arevalo focuses on ensuring new capabilities strengthen, rather than undermine, the clinical judgment and human relationships at the center of behavioral health and IDD care.
No comments:
Post a Comment