Which AI Tools Can a Billing Company Use With Patient Data? A 2026 BAA Checklist

Your billers are probably already using a chatbot. The question is whether it is one your firm can legally put patient data into.

This is a practical checklist for billing-company owners, current as of October 2026. It is not legal advice. Have your compliance lead or attorney confirm anything you rely on, and check each vendor's current terms yourself, because they change.

The rule in one paragraph

Under HIPAA (the Health Insurance Portability and Accountability Act), a billing company is a business associate of the practices it serves. When you hand protected health information (PHI) to another vendor to do work for you, that vendor needs a business associate agreement (BAA) with you. An AI tool is a vendor. If there is no BAA covering the specific product and plan your staff use, patient data does not go into it.

Which AI plans offer a BAA (as of October 2026)

Anthropic (Claude)

According to Anthropic's support page (accessed Oct 2026):

• Claude Enterprise can be covered, once the HIPAA setting is turned on and the BAA is accepted by the account's Primary Owner.
• The Anthropic API can be covered after a BAA is signed and PHI use is enabled through Anthropic's sales team.
• Claude Code and Cowork in local mode are covered only under a HIPAA-configured Enterprise plan.
• Not covered: the Claude Console, cloud Cowork, and beta features such as Claude Docs, Slides and Design.

That last line matters. Even on a covered plan, some features are outside the BAA. Your policy should name them.

OpenAI (ChatGPT)

According to a third-party summary by IntuitionLabs (Jul 2026, updated Aug 2026), which you should confirm with OpenAI:

• ChatGPT Free, Go and Plus do not come with a BAA.
• Self-serve ChatGPT Business ($25 per user per month billed monthly, $20 billed annually) does not come with a BAA.
• ChatGPT Enterprise: a BAA is available only for sales-managed accounts.
• ChatGPT for Healthcare offers a BAA at custom pricing.

The trap here is the word "Business." A paid business plan is not automatically a HIPAA-eligible plan.

Google and others

We have not verified Google's Gemini terms for this post. Check directly. The same goes for any AI feature built into your practice-management system or clearinghouse: ask whether that feature is inside the vendor's existing BAA with you.

AI phone and voice vendors

Several vendors sell AI agents that call payers for claim status. Pricing is rarely published (Prosper AI, Oct 2026, vendor article). Ask each one for its BAA and its security reports before a pilot. One example: SuperDial says it completed a SOC 2 Type II audit in November 2025 (SuperDial, Jul 2026, vendor claim). A SOC 2 report is not a BAA; you need both questions answered.

The one-page policy

Every billing company using AI should have a short written policy. Here is what goes on it:

1. Approved tools list. Product name, plan, and whether it is covered by a BAA. Anything not on the list is not allowed with patient data.
2. What may go where. For example: claim and denial data only in approved tools; de-identified text only in anything else.
3. Minimum necessary. Each task gets only the fields it needs. An appeal drafter does not need a patient's full history.
4. Templates are de-identified. Prompt libraries and appeal templates never contain real patient details.
5. A person approves every outbound item. No appeal, status note or client message goes out without a biller's review.
6. No code changes, no claim submission by AI. It can flag. People decide.
7. Features outside the BAA are named. For example, beta features on an otherwise covered plan.
8. Accounts are company-owned. No personal logins. Remove access the day someone leaves.
9. Incident steps. Who to tell if PHI goes into the wrong tool, and how fast.
10. Review date. Re-check vendor terms at least twice a year.

Questions to ask any AI vendor

• Will you sign a BAA, and which exact products and plans does it cover?
• Are any features excluded from the BAA?
• Is our data used to train your models? Can we turn that off in writing?
• Where is data stored, and for how long?
• Who on your side can see our data, and is that logged?
• What security audits have you completed, and can we read the report?
• How do we export or delete our data if we leave?

Common ways firms get this wrong

• Personal chatbot accounts. A biller pastes an EOB (explanation of benefits) into a personal Plus account to "just summarize it." That is the most likely leak in most firms.
• Browser extensions. AI writing helpers that read every page, including the payer portal.
• Screenshots. Uploading a screenshot of a claim screen is still uploading PHI.
• Assuming the paid plan is covered. As above, "Business" is not the same as "BAA."
• Assuming the whole product is covered. Covered plans can still exclude beta features.

Why this matters more in 2026

Payers are deep into AI themselves: more than half of health plans use AI tools in administrative work, versus 25% of provider organizations (CAQH Index via Dataspring, Feb 2026). Billing firms will adopt AI too. The firms that can show a client a written policy, a list of covered tools and a person approving every outbound item will have an easier renewal conversation than the ones that cannot.

Get the template

Lunan has a free one-page PHI-safe AI tool policy template for billing companies, built on the list above. Fill in your tools and your rules, and have your compliance lead review it.

If you want help choosing covered tools and setting up the first workflow on them, start a free conversation. We will tell you where patient data would go before anything is built.