Texas Would Make AI a Licensure Matter for Behavior Analysts. It Hasn't Adopted It.

TDLR's proposed 16 TAC §121.100 bars a Texas LBA from letting AI be the sole basis of a treatment design, assessment or report, and new Code of Ethics items require written client consent before client data goes into an AI tool — including for report generation. Comments closed July 20. Nothing has been adopted.

In January, Texas HB 149 gave providers an AI disclosure duty. The Texas Department of Licensing and Regulation has proposed something further downstream: clinical-practice duties written into the behavior-analyst licence itself, enforceable through the board's own disciplinary process.

The proposal was filed with the Secretary of State on June 8, 2026 and published in the Texas Register on June 19. Comments closed in July. As of today the Commission has not adopted it, and there is no effective date.

What it would require

New Subchapter H, §121.100, in six subsections. The first sets the rule:

“A license holder must not use AI as the sole basis for a client treatment design, assessment, treatment implementation, report, or for the evaluation of a treatment plan.”

Then the paperwork. A licence holder must individually review anything “generated in whole or in part by AI.” Any AI-derived change to a plan or assessment “must be based on accurate evidence personally verified by the license holder, and the license holder must document this verification in the client's file.” The licence holder must also be able to “explain to a client the evidentiary basis for these changes in understandable language” — and document that explanation in the file too.

Restrictive procedures get their own subsection. Where an AI-generated product recommends a restrictive or punishment-based procedure, the licence holder must personally confirm less intrusive means would not work and that benefit outweighs risk of harm, and document the rationale in the file.

The consent clause is the one to read twice

Separately, §121.75(b)(12) adds to the Code of Ethics a duty to obtain written client consent before using the client's confidential or personal data for “client treatment design, assessment, treatment implementation, report generation, usage in marketing materials, or the evaluation of a treatment plan” with AI — and separately for providing that data to train AI.

Report generation is in the list. If a practice uses an AI tool to draft session notes or progress reports, this is a consent requirement covering routine documentation, not an edge case about model training.

Two more items follow. Where the AI is “controlled by the license holder's employer,” the licence holder must record in the file the date a copy of the consent went to the employer — a duty that lands on the individual, not the clinic. And any known unauthorized disclosure of protected health information triggers a three-part duty: record who received it, notify the client, keep proof of the notification.

The limits

None of this is in force. It binds nobody today, and TDLR adopted six other programs' rule packages on August 31 and September 1 without this one. “AI” is not TDLR's own definition — §121.10 imports the HB 149 meaning from Business and Commerce Code §551.001(1), which is broad enough to cover a scheduling optimizer.

The duties fall on individual LBAs and LaBAs, not on the clinic as an entity and not on behavior technicians, who are not licensed in Texas. A record-retention amendment that had been drafted was removed before publication; retention stays at six years. And TDLR's fiscal note asserts compliance is free: “License holders should incur no costs to comply with the AI ethics requirements…”

What you must know or do

  • Answer one question this week: does any client data currently go into an AI tool at your practice, and is there written consent on file for it? If you draft notes or reports with AI, under this proposal the answer needs to be yes before it starts, not after.
  • Check who owns the tool. Where the employer controls it, the consent-copy log is the individual licensee's duty. A clinic-level policy does not discharge it.
  • Look at whether your documentation has a place to record a verification. Two of the six subsections require a note in the client's file that a named licensee personally checked an AI-derived change. Most templates have nowhere to put that.
  • Watch the TDLR Commission's adoption agenda. Adoption is what starts the clock, and until it happens the text can still change.