Claude for Healthcare Is Already in Some EHRs. Why Is Yours Still in Pilot?

What separates the health systems that shipped from the ones still stuck.

When Anthropic launched Claude for Healthcare at the J.P. Morgan Healthcare Conference (JPM) in January 2026, at least one Electronic Health Record vendor had it embedded into their system within the same week. Most health system CIOs read that headline and had the same reaction: how did they do that so fast?

The answer wasn’t the AI. It was everything that vendor had built before the AI arrived — a working data infrastructure, a documented compliance framework, and an integration layer that was ready before Claude entered the picture.

Most organizations don’t have that environment. And that’s why most of them, months after signing their agreement with Anthropic, are still running a pilot.

What Anthropic Actually Launched

Claude for Healthcare is a dedicated product layer built by Anthropic for hospitals, health systems, and payers. It includes:

• A compliance agreement (Business Associate Agreement) that legally allows patient health data to flow through the AI
• Pre-built connections to Medicare coverage rules, medical diagnosis codes, and provider registries
• Tools for developers to build healthcare integrations using the industry standard data format (HL7 FHIR)
• Support for prior authorization checks and claims appeal workflows

This is a real product with real deployments. The capability is there. What it does not include is the integration work that makes it functional inside your specific systems.

The Agreement Is Not the Integration

Signing the compliance agreement with Anthropic is the right first step. But it’s a legal permission, not a technical solution.

After the signature, the real questions start:

• Where does AI output go — into your Electronic Health Record, or into a separate tool that clinicians won’t actually use?
• What happens when the AI summarizes a patient record using data from yesterday, and the patient had new orders placed this morning?
• Who reviews what the AI generates before it becomes part of a clinical record?
• If that document is ever requested in a legal proceeding, is there a clear log of how it was created?

These are not edge cases. They determine whether an AI deployment works in the real world or stays permanently in pilot. And none of them are answered by the compliance agreement.

The organizations that move from signed agreement to live production in weeks rather than months aren’t faster because they have more budget or better developers. They’re faster because they had already answered those questions before the AI conversation started.

What You Need Before the AI Can Work

What needs to exist	Why it matters
Real-time data connection to your EHR	AI that reasons over yesterday’s patient data is not a clinical tool — it’s a liability. The connection needs to be live, not a nightly data export.
Authenticated access controls	The AI should only see what the logged-in clinician is authorized to see. Without this, you have a compliance exposure waiting to happen.
Write-back into the EHR	An AI-generated note that lives only in the AI platform is not a clinical record. It needs to exist inside your EHR — timestamped, versioned, attributed correctly.
A full audit trail	Every AI-assisted document needs a log: what data was used, when, what a human reviewed, what changed. This is what compliance and legal will ask for.

The health systems that deployed Claude for Healthcare into production — not pilot — had all four of these in place before AI entered the conversation. The integration layer was already there, and the AI had somewhere to land.

The Cost of Skipping the Foundation

The pressure to move fast is real. But moving fast without the foundation creates a different kind of problem.

An AI-generated clinical note without documented clinician review is a liability. Patient data flowing through an AI system without proper logging is a compliance gap. A prior authorization workflow running through AI without a structured handoff to the billing system saves no one any time.

Pilots stay in pilot for a reason, and it’s usually the same one: the AI layer got built before the foundation underneath it was ready, and now the team is working backwards through problems that should have been solved first.

What Itirra Builds

We don’t sell AI. We build the infrastructure that makes AI work inside clinical systems.

That means the data connections, the security and authentication layer, the write-back integrations, and the audit logging — the foundation that determines whether AI becomes a real clinical tool or a permanent pilot.

For health systems evaluating Claude for Healthcare: the right first question isn’t which use case to start with. It’s whether your current infrastructure supports real-time data exchange, authenticated access, and a defensible audit trail. If the answer is unclear, that’s where the work begins.

---

Claude for Healthcare is already in some EHRs. The difference between those health systems and everyone else isn’t budget, and it isn’t technology access. It’s the integration layer that was there before the AI arrived. This is the entry point — not the finish line. The organizations that get that right now will have a very different set of options in 12 months.

#HealthcareIT #AIinHealthcare #EHRIntegration #ClaudeForHealthcare #HIPAA #FHIR