For as long as I have been building agentic AI systems, I have heard a version of the same objection when the conversation turns to healthcare. Protected health information cannot touch an AI agent, they say. The liability is too high, the compliance burden too strict, the margin for a confident wrong answer too thin. A major U.S. health system just gave the enterprise agentic AI community a clean counter example, in one of the most administratively burdened corners of medicine: prior authorization.

Prior authorization is the process where a clinician has to justify a treatment to a payer before insurance will cover it. It routinely takes days, sometimes longer, and it delays care patients are already waiting for. It has resisted automation for years because it sits at the intersection of clinical judgment, payer policy and protected health data, exactly the kind of intersection skeptics point to when they say agentic AI does not belong in regulated healthcare. I want to walk through what happened, in the words of five credible sources, then explain it using my own roadmap.

What Happened

A HIPAA-ready agentic AI platform was rolled out at a large U.S. health system to help clinicians and administrative staff handle prior authorization and related documentation review. The system connects to coverage databases and clinical records to check a request against payer policy, assemble the supporting evidence, and propose a determination for human review before anything is submitted to a payer. In one pilot, the platform worked through more than 1,400 pages of clinical notes, and the large majority of clinicians using it reported meaningful time savings with no drop in accuracy. The aim, as one industry executive close to the rollout put it, is to take repetitive paper work off a clinician’s routine so more time goes back to the patient in front of them.

Here is what five credible sources covered this news story:

TechCrunch covered the underlying platform launch, reporting that it lets clinicians speed up prior authorization review, the process where a doctor submits additional information to an insurer to confirm coverage for a treatment. Read the TechCrunch coverage.

Fierce Healthcare reported that the HIPAA-ready infrastructure behind the platform allows healthcare organizations to deploy agentic AI on workflows involving protected health information, including prior authorization, for the first time. Read the Fierce Healthcare coverage.

Becker’s Hospital Review detailed how the tool pulls coverage requirements from payer or Medicare policy, checks clinical criteria against patient records in a HIPAA-ready manner, and proposes a determination with supporting materials for the payer’s review. Read the Becker’s coverage.

2 Minute Medicine framed the rollout as governed workflow software rather than a consumer chatbot, and noted that hospitals will still need continuous monitoring and clear escalation paths so a confident AI error never quietly enters a real clinical process. Read the 2 Minute Medicine coverage.

Anthropic’s own release stated the tools were built so that patients could get life-saving care more quickly and so clinicians could spend less time on paperwork and more time on patients. Read the original announcement.

Running It Through The Definition

My definition of agentic AI is simple. Intent in, outcomes out. You state the goal, the system plans and executes the steps, and a human stays accountable for what happens at the end. This case study fits that definition exactly. The clinician’s intent is a covered treatment for a patient who needs it. The agent plans the multi-step path to get there, checking policy, checking records, assembling evidence, drafting the request. The outcome is a proposed determination, not a final one, because a human still signs off before anything reaches the payer.

Against the Five-Stage Agentic AI Roadmap, Prompted, Piloted, Governed, Assured, Autonomous, I place this case study at Stage 3 moving toward Stage 4. It is Governed because every proposed determination routes through a human before submission, and the system is built on connectors to authoritative data sources rather than free-form generation. It is moving toward Assured because the health system is running this at meaningful volume, with clinician-reported outcomes now public, which is exactly the evidence base that assurance requires before an organization extends autonomy further.

My conclusion for this case study mirrors the one I have drawn from other regulated case studies. If governance and human accountability are built in from day one, agentic AI belongs in healthcare’s most sensitive administrative workflows, not despite the stakes, but because the stakes are exactly what governance is designed for.

The naysayers were half right. Protected health information does demand a higher bar. What they missed is that the bar has already been cleared.

Dr. Harish Kotadia, Ph.D.

Disclaimer: The content of this blog post is drawn from reputed media sources available in the public domain and linked above. This post is intended for educational purposes, to help the enterprise agentic AI community learn from public information on the application and use of agentic AI tools and technology in Fortune 500 companies. The views and opinions expressed here are my own and do not represent those of any employer or client, past or present. The analysis presented is my independent interpretation of published reporting and does not constitute legal, financial, or consulting advice of any kind.

© Dr. Harish Kotadia, Ph.D. All Rights Reserved, 2026

Dr. Harish Kotadia, Ph.D., is an Enterprise AI Architect with 20+ years of IT consulting experience serving Fortune 100 clients, specializing in agentic AI systems built on Anthropic Claude, AWS Bedrock, and Google Vertex AI. Views expressed are his own. Follow at www.linkedin.com/in/hkotadia and AgenticAIArch.com.