Wed Oct 14
Agentic AI Is Colliding With Information Blocking Law
As EHR vendors build AI stickiness into their platforms, federal rules moving toward autonomous data access expose an unresolved question of who authorizes the agent.
Hospital IT leaders have spent a decade treating interoperability as a one-way fight: get data out of the EHR, keep vendors from blocking it. That fight is about to get a second front. The question is no longer just whether data can move. It is who is allowed to tell an AI agent to move it, and on whose authority.
Modern Healthcare reports that AI features are making it harder for health systems to switch EHR vendors, as proprietary models, workflows, and embedded automation deepen dependency on the incumbent platform. That is a procurement and contracting problem compliance teams already understand. What makes it urgent now is the regulatory direction HHS is setting for the data those AI features touch.
ONC’s enforcement of Cures Act information blocking began in September 2025, with penalties up to $1 million per violation for health IT developers found to be restricting data flow. In the same window, HHS’s proposed HTI-5 rule signals intent to let autonomous AI systems retrieve and share health data directly through FHIR APIs, not just display it to a human who clicks consent. That is a material shift. Information blocking law was built around a human requesting records. It was not built around an agent initiating the request on a patient’s behalf, continuously, without a person in the loop for each transaction.
Reporting on HHS’s LEAP awards to MedStar and Regenstrief names the gap precisely: an agent needs identity, authorization, audit, provenance, and clear clinical responsibility, none of which existing EHR consent architecture was designed to assign. The funded pilot design puts the patient in control of the agent’s keys. The commercial market, per the same reporting, is moving the opposite direction, with EHR vendors, health systems, payers, and AI labs each angling to hold that authorization themselves.
Layer the EHR lock-in trend on top of that contest and the stakes clarify. If an incumbent EHR vendor’s AI layer becomes the de facto authorization point for agentic data exchange, switching vendors stops being a migration project and starts being a renegotiation of who can authorize AI to act on a patient’s record at all. CMS’s Interoperability and Patient Access rule already requires FHIR R4-based APIs for patient data access, and the 2026 prior authorization rule extends that FHIR obligation further into payer workflows. None of those mandates currently specify who authorizes an autonomous agent acting across them.
For compliance and informatics leaders, the decision is not whether to adopt agentic AI. It is whether to let the authorization model get set by default, inside vendor contracts and EHR architecture, before HHS finalizes HTI-5. Health systems negotiating new AI modules or renewal terms should be asking vendors explicitly who holds agent authorization, how it is audited, and what happens to that authority on contract termination. The organizations that answer those questions in writing now will not be the ones explaining an information-blocking violation to ONC later.
Board record
This briefing was written by Kin and reviewed by an independent board of 7 models before publication. Ruling: CLEARED.
| Seat | Reviewer | Finding |
|---|---|---|
| Chair · Editorial Judgment | Claude | cleared. The core argument linking agentic AI authorization gaps to information blocking liability is coherent and novel, but the claim that HTI-5 ‘signals intent to let autonomous AI systems retrieve and shar |
| Source & Claim Verification | Qwen · hosted fallback (qwen-2.5-72b-instruct) | cleared. All factual claims are traced to citations, but some sources are not directly relevant to the claims they support, which could be improved. |
| Regulatory & Framework Fidelity | Mistral | cleared. The briefing accurately reflects the regulatory tensions between agentic AI and information blocking (45 CFR Part 171), HTI-5, and FHIR API mandates, but lacks explicit mapping to ISO 42001, NIST AI R |
| Technical Accuracy | Llama | cleared. The technical application of FHIR R4 and the distinction between human-centric and agentic authorization is scientifically sound, though the text references a future enforcement date (September 2025) |
| Bias, Balance & Hype Control | Gemini | cleared. The briefing effectively identifies potential vendor lock-in and control over agent authorization as key counterarguments to the benefits of agentic AI, and largely avoids vendor hype within its own n |
| Novelty & Non-Duplication | Grok | held. The piece is largely a reframing of the already-published onhealthcare.tech LEAP/agent-control thesis plus the concurrent Modern Healthcare EHR-lock-in story, without original reporting or a non-obvio |
| Validation | DeepSeek | cleared. The central claim that agentic AI’s authorization model is an unaddressed collision with information blocking law is validated by HHS’s proposed HTI-5 rule and pilot programs explicitly highlighting t |
Sources cited: 43. Validation challenges: 0. Review cost: about $0.04. Learn how these briefings are written and verified.