Is Claude HIPAA Compliant? BAA Paths and Pricing

"Can we use Claude with patient data?" has a more textured answer than most vendors give, because Anthropic has drawn the boundary carefully and written it down.

The short answer

Yes, on specific products. Anthropic signs a BAA covering its first-party API and Claude Enterprise, and for qualified accounts Claude Code with zero data retention enabled. The BAA explicitly does not cover consumer plans.

From Anthropic's Trust Center:

After review of HIPAA-related compliance items and the customer's use case, Anthropic may provide a Business Associate Agreement (BAA) covering use of eligible products. Our BAA only covers eligible products defined in the Implementation Guide for HIPAA Entities, including Claude Enterprise Chat, Anthropic's native 1P API, and (for qualified accounts) Claude Code with Zero Data Retention (ZDR) enabled. Our BAA does not cover beta products or Free, Pro, Max, or Team plans.

Two things stand out. Acceptance follows a review rather than being self-serve for everyone, and the exclusion list is explicit rather than implied.

What the BAA does not cover

Anthropic's documentation is unusually candid here, and the list is worth reading before you architect around it:

• Claude Free, Pro, and Max - the consumer plans.
• The Claude Console, including the playground.
• Partner-operated platforms - Amazon Bedrock and Google Cloud's Agent Platform. Anthropic notes that on those platforms the cloud provider is the data processor.
• Claude Platform on AWS and Microsoft Foundry, where HIPAA readiness is not available at all.
• Third-party integrations connected to your application.
• Claude Code, which is not covered under HIPAA readiness unless very specific criteria are met.
• Beta features, unless explicitly listed as eligible.

That partner-platform point matters. If you are running Claude through Bedrock, you are not relying on Anthropic's BAA. You are relying on AWS's, because Bedrock is a HIPAA-eligible AWS service. Anthropic states the boundary directly: Claude Code purchased "through Amazon Bedrock, Google Cloud's Agent Platform, or Microsoft Foundry" is "available but not covered under BAA."

How to get the agreement

There are two paths, and they differ in effort.

Claude Enterprise. The Primary Owner activates HIPAA compliance in the organization settings under "Data and privacy" and accepts the BAA. Anthropic notes that standard Enterprise plans "do not include BAA coverage without action from a Primary Owner" - it is not automatic.

First-party API. Anthropic's instruction is to sign the BAA and then contact your account team or sales to have it enabled: "To use the 1P API with PHI, your organization's Primary Owner will need to sign a BAA and then reach out to your Anthropic contact or our Sales team to get this turned on."

HIPAA readiness versus zero data retention

Anthropic separates these two things more clearly than most providers, and the distinction is useful.

Their platform documentation notes that HIPAA readiness "applies a broader set of privacy and security safeguards than ZDR (encryption, access controls, and audit logging that protect PHI throughout its lifecycle) rather than requiring immediate deletion." It then adds a sentence worth reading twice: "If your organization handles PHI, HIPAA readiness is the arrangement to use; you do not also need ZDR."

That is the opposite of the usual vendor framing, where ZDR is presented as the premium tier. Anthropic treats PHI handling as a configuration with safeguards, not simply as a deletion setting.

What happens to the data

• Default API retention. Inputs and outputs are deleted on Anthropic's backend "within 30 days of receipt or generation," with exceptions for longer-retention services you enable, agreed arrangements, Usage Policy enforcement, and legal requirements.
• Policy violations. If a chat is flagged by automated trust and safety systems, "we retain inputs and outputs for up to 2 years and trust and safety classification scores for up to 7 years."
• Covered models. Some models - Anthropic designates them "Mythos-class" - carry a 30-day retention requirement "on every platform where these models are offered," and are unavailable under ZDR unless expressly authorized.
• Training. The Commercial Terms state: "Anthropic may not train models on Customer Content from Services." Retained data is "never used for model training without your express permission."

Pricing

Representative Claude API prices at the time of writing, per million tokens:

Two pricing notes from Anthropic's own documentation. Sonnet 5's introductory $2/$10 pricing "is now the standard price," and the previously scheduled increase to $3/$15 "will not occur." The Batch API carries a 50% discount on both input and output tokens.

For comparison, current open-weight rates on our catalog run from $0.14/M input for DeepSeek-V4-Flash-Vision-Exp to $3.00/M input for Kimi K3, with Sonnet-class capability available at a fraction of Sonnet 5's price on several of the newer open checkpoints.

Where Claude is the right choice

• You need Claude specifically. Frontier reasoning, long documents, and its writing quality are real advantages, and we do not serve closed models today.
• You are already on Claude Enterprise. In that case the BAA is a settings change away.
• Your workload never touches PHI. Then the consumer and beta exclusions are irrelevant.

Where a BAA-backed open-weight API is the right choice

• Anything clinical. At that point the exclusion list above becomes your architecture constraint.
• Cost at volume. A 10x per-token difference is a rounding error at 10 million tokens and a budget line at 10 billion.
• Model portability. Open weights can move between providers. A hosted closed model cannot.
• One compliance surface. One BAA, one retention posture, one audit trail, rather than a partner platform's agreement layered on a model vendor's.

The decision rule

Ask which product surface the workload runs on. Claude via Anthropic's own API or Claude Enterprise can sit under an Anthropic BAA, subject to review. Claude via Bedrock sits under AWS's. Claude via the Console or a consumer plan cannot carry PHI at all.

Then ask whether you need a frontier closed model for the task. For summarization, extraction, classification, and drafting, an open weight usually closes most of the gap at a fraction of the price - and the HIPAA compliance guide covers the agreement you need on either path.

OpenMed Router provides a signed BAA, zero-retention isolated inference, and AES-256 encryption across the open-weight model catalog. Proprietary models are coming soon.

Weighing Claude against open weights? Join the waitlist.