Is Claude HIPAA Compliant? Every Plan, Checked

Short answer: Claude can be used with protected health information only on a HIPAA-ready Claude Enterprise plan, with HIPAA enabled and the Business Associate Agreement accepted by the organization's Primary Owner, or through Anthropic's API under a signed BAA. Anthropic's help center is blunt about the rest: "Team plans and individual plans (Free, Pro, Max, and Ultra) can't enable HIPAA."

This post is for the internal IT lead or compliance officer who has just realized the question is not hypothetical. In most of the firms we talk to, nobody has formally adopted AI yet, and some staff are already using Claude or ChatGPT on personal accounts to draft letters, summarize documents and clean up spreadsheets. That is how email and smartphones arrived at work too: people used them first and policy caught up later. The job now is to move that work onto a plan the company controls, and to know exactly what that plan covers.

Claude HIPAA eligibility, plan by plan

From Anthropic's help and privacy centers as served on November 2, 2026. Anthropic edits these pages often (both were marked "Updated this week"), so check them again before you sign anything.

Two lines from Anthropic's BAA page for commercial customers do most of the work here. First: "Standard Claude Enterprise plans do not include BAA coverage without action from a Primary Owner." Buying Enterprise does not switch HIPAA on; someone has to do it deliberately. Second: "the BAA only covers the single organization that accepted it." A department that opened its own Claude organization is not covered by the BAA your main organization accepted.

What a personal Claude account does with work data

The risk with personal accounts is not that Claude is careless. The problem is that a personal account is a consumer contract between your employee and Anthropic. Your company is not a party to it and has no admin console for it. Anthropic's August 28, 2025 consumer-terms update set the current rules for Free, Pro and Max:

• Training is a user setting. Anthropic will "train new models using data from Free, Pro, and Max accounts when this setting is on." Users had until October 8, 2025 to choose.
• Retention depends on that choice. "We are also extending data retention to five years, if you allow us to use your data for model training." Otherwise: "you'll continue with our existing 30-day data retention period."
• Business plans are excluded. The update does "not apply to services under our Commercial Terms, including: Claude for Work, which includes our Team and Enterprise plans" and the API, including through Amazon Bedrock and Google Cloud.

On the business side, Anthropic's Commercial Terms say: "Anthropic may not train models on Customer Content from Services." There is one exception to brief staff on. Anthropic's privacy center says that if someone explicitly reports feedback, for example with the thumbs up/down button, Anthropic may use that chat for training, even on commercial plans. Tell people not to rate conversations that hold client or patient data.

On a personal account, the training setting belongs to the employee, not to you. Whether a given client file went into five-year retention depends on a checkbox you cannot see. Moving staff to Team or Enterprise does not fix HIPAA by itself, but it does put the data under a contract your company signed, with an admin who can see the organization.

Enterprise with a BAA: what is covered and what is not

Turning on HIPAA does not cover every Claude feature. Anthropic's HIPAA-ready Enterprise article puts it this way: "enabling HIPAA doesn't bring every feature under your BAA. Features fall into three categories: covered by your BAA, available but not covered, and disabled." From the BAA page's coverage table:

• Covered: chat; projects in chat; artifacts in chat; file creation and code execution (excluding network access and external websites); voice; web search; research.
• Available but not covered: MCP connectors, Enterprise Search ("Ask Your Org") and Claude in Chrome, where data flows to third parties; Skills; Cowork in the cloud; parts of Claude for Microsoft 365 that are in beta.
• Not available to HIPAA-ready organizations yet: Claude Design, Claude Slides and Claude Docs (all beta).
• Claude Code and Cowork on the desktop are covered only after the Primary Owner applies a separate "HIPAA configuration," which Anthropic says is "currently only available to organizations with an Anthropic account team." Claude Code reached through a cloud provider or an LLM gateway is not covered.

The admin work here belongs to IT, not legal. An Enterprise plan with HIPAA enabled and connectors left on still gives users a path to send PHI to an uncovered feature. Anthropic's instruction is clear: "Process PHI only with a feature marked ✅ Covered." Your configuration has to make that the default rather than leave it to each user.

Claude may already be in your Microsoft tenant

A separate route: Microsoft enables Anthropic models inside Microsoft 365 Copilot for most commercial tenants, with Anthropic acting as a Microsoft subprocessor. That route runs under your Microsoft agreement, not an Anthropic plan, so the HIPAA question there goes to Microsoft's BAA and your tenant settings. We covered the setting and its limits in Copilot, GCC and the Anthropic boundary and the Microsoft BAA in Is Microsoft Copilot HIPAA compliant?

What internal IT should do this month

1. Find out what is already in use. Check proxy, firewall or DNS logs, or Defender for Cloud Apps discovery if you license it, for claude.ai, chatgpt.com, gemini.google.com and similar. Count users, not just hits. You are sizing the problem, not hunting individuals.
2. Pick the sanctioned tool. If the firm handles PHI and staff prefer Claude, the eligible path is Enterprise with HIPAA enabled. If Team is the budget, it is fine for non-PHI work, but the written rule has to say that PHI never goes into it.
3. Turn it on deliberately. Find out who your Primary Owner is, enable HIPAA, and review the uncovered features (connectors, Chrome, Skills) before users find them. Anthropic notes that enabling HIPAA "resets certain settings across your organization," so do it before you configure the rest, not after.
4. Write the one-page rule. Approved tools, what data never goes in, who to ask. Our AI usage review includes an acceptable-use policy draft; your privacy officer and counsel own the final version.
5. Close the old accounts. Ask staff to move work out of personal accounts, delete conversations that held client or patient data, and switch off the training setting on anything they keep.

None of this is legal advice. Whether a specific past disclosure is a reportable breach is a HIPAA determination for your privacy officer. The technical controls above are what IT can put in place this month.