ChatGPT and Claude in Your ABA Clinic: Which Plans Can Touch Client Data
Informational only; not legal, compliance or clinical advice. Vendor plan details described as of Nov. 2, 2026, based on each vendor's own help center.
General AI assistants have quietly moved into ABA. Flychain, a bookkeeping firm for healthcare practices, looked at P&L data from more than 100 ABA practices and reported that, among practices on its books since January 2025, the share paying for at least one ChatGPT or Claude subscription rose from 19% in Q1 2025 to 50% in Q1 2026. The median spend among those practices was about $80 a month, which Flychain reads as one or two individual seats rather than a clinic-wide rollout.
That pattern matters. A $20 seat bought on someone's personal card is very different from a plan the clinic set up to handle protected health information. Before a BCBA pastes a session summary into a chat window, it is worth knowing which is which.
What OpenAI says is covered
OpenAI's help center lists the products it makes available with a business associate agreement (BAA): ChatGPT for Healthcare, ChatGPT for Enterprise with Regulated Workspace, ChatGPT FedRAMP, ChatGPT for Clinicians, and the API with Modified Retention (plus a FedRAMP API version).
Two details are easy to miss:
- Not every feature is covered, even on those plans. OpenAI lists features that are covered when used from an eligible workspace, and a separate list that is not covered, including improved memory and event-triggered scheduled tasks. It says PHI should not be entered when using those features.
- Features not on the list are not automatically covered. OpenAI notes that recently added functionality may not yet be listed, so "it's available in our workspace" is not the same as "it's under our BAA."
Individual and small-team ChatGPT subscriptions, such as Plus and Business, do not appear on that eligible list.
What Anthropic says is covered
Anthropic's help center says its HIPAA-ready offering is available on Claude Enterprise plans, with a BAA that eligible organizations can accept in organization settings. It states plainly that Team plans and individual plans (Free, Pro, Max and Ultra) can't enable HIPAA.
Anthropic says enabling HIPAA "doesn't bring every feature under your BAA," sorts features into covered, available but not covered, and disabled, and points administrators to its Implementation Guide as the authoritative list. It also describes the switch as a one-way decision that resets some settings, and says only the organization's Primary Owner can accept the BAA.
The practical takeaway: two lanes
Most clinics don't need an enterprise contract to get real value from these tools. They need a clear rule about which lane each task belongs in.
Lane one: no client information at all. This is where a regular subscription is useful. Good fits:
- Drafting a caregiver handout on a general topic, like building a bedtime routine.
- Writing job posts, interview questions and onboarding checklists for RBTs.
- Turning a payer's published policy into a plain-language checklist for your team.
- Drafting internal policies, such as your clinic's own AI use policy.
Lane two: anything about a specific learner. Session notes, treatment plan language, reauthorization narratives, caregiver emails about a particular child. These belong in a tool your clinic has a BAA with and has configured, whether that's a covered ChatGPT or Claude plan or an ABA platform whose AI features fall under your agreement.
"I'll just remove the name" is not enough
HHS guidance describes two ways to de-identify health information. Under the Safe Harbor method, a long list of identifiers must be removed, including names, dates more specific than the year, geographic details smaller than a state, phone numbers, emails and record numbers. Initials don't qualify. The guidance also says identifiers must be removed from free-text fields too, and that the information isn't de-identified if you know the rest could still identify someone.
In ABA that bar is easy to trip. A note about "a 4-year-old at our Tuesday clinic who just started a new school" may identify a family without a single name in it. If a task needs learner details to be useful, treat it as lane two.
What to do this week
- Ask your team which AI accounts they use for work, and whose card pays for them.
- Write down the two lanes and share them in your next staff meeting.
- If you want AI help with learner-specific work, pick one covered tool, sign the BAA, and turn off the features the vendor says aren't covered.
- Keep the BCBA in charge of every clinical word that reaches the record.
AI should give clinicians back time for the work that changes outcomes. A one-page rule about where client information goes is what lets you use it without second-guessing.
Browse the ABA AI Tools directory to compare general assistants and ABA-specific documentation tools.
Sources
- https://www.flychain.us/resources/2026-aba-ai-adoption-spend-report
- https://help.openai.com/en/articles/20001069-hipaa-eligible-products-and-functionality
- https://support.claude.com/en/articles/13296973-hipaa-ready-enterprise-plans
- https://www.hhs.gov/hipaa/for-professionals/privacy/special-topics/de-identification/index.html
More from the blog
- A One-Page AI Use Policy for Your ABA Clinic: What to Put In It
- AI Meeting Note-Takers in ABA Clinics: What to Check Before the Bot Joins
- SOC 2, BAAs and Badges: How to Read an AI Vendor's Security Claims Before Your ABA Clinic Signs
- An Autism Center Replaced Its Supervisor With AI. Here Is What AI Can't Take Over in RBT Supervision
Informational only; not legal, clinical or compliance advice. Vendor HIPAA/BAA statements are as published by the vendors and are not verified by us. Sponsorship never changes our rankings. Corrections: admin@gtbehaviorbolts.com.