Kansas addiction-counseling advisory committee opens discussion on regulating artificial intelligence in practice

Jill Flesing, chair of the Addiction Counselor Advisory Committee of the Kansas Behavioral Sciences Regulatory Board, opened an advisory-committee discussion on regulation of artificial intelligence in behavioral-health practice during an otherwise routine remote meeting.

David Fye, executive director of the Kansas Behavioral Sciences Regulatory Board, framed the packet of materials the board provided to advisory committees: a state executive-branch generative-AI policy from the Office of Information Technology Services (OITS), guidance from the American Association of State Counseling Boards, and a Kansas Health Institute template for AI policies in public-health organizations. Fye said the OITS document includes a provision that “responses generated from generative AI shall not be used verbatim, be assumed to be truthful, . . . be used to issue official statements, [or] be solely relied upon for making final decisions.”

The committee focused on several practical risks: recordings and transcripts created for AI-assisted documentation, retention and subpoena risk for session audio, the ability of AI to “hallucinate” inaccurate material, and the difficulty of identifying speakers in group sessions. Michael James, an advisory committee member who participated in a vendor demonstration, said the EHR integrations he saw generated a transcript from recorded sessions and used those transcripts to draft clinical notes and treatment plans. He raised questions about how long recordings would be stored, whether they become official medical records subject to subpoena, and whether systems would track what percentage of AI-generated text clinicians altered before signing.

Members described a range of current uses. Some universities and clinics already use AI-generated summaries, templates and wording for nonclinical documents; several members said providers and students are experimenting with AI for drafting but that clinical leaders expect human review. Committee members stressed that any permitted use should include training, human oversight and clear patient informed-consent language.

Fye identified three documents the board provided for review: the state agency generative-AI policy (to protect privacy, intellectual property and safety), the American Association of State Counseling Boards guidance (which recommends informed consent, data safeguards and opt-out rights for clients), and a 76-page Kansas Health Institute/Wichita State University policy template (which emphasizes human oversight, transparency and data governance). He asked the advisory committee to review those materials and said the board would welcome recommendations on whether the BSRB should set standards for acceptable AI tools, use-limits or informed-consent language.

Members suggested concrete topics for any future guidance: whether and how recorded session audio used by AI becomes a part of the medical record; minimum human-review standards for AI-generated clinical notes; limits on the use of open-source public models that retain input data; informed-consent language allowing clients to opt out of AI-assisted care; and training or continuing-education requirements for clinicians who employ AI in clinical work.

The committee agreed to continue the discussion at future meetings and to watch an educational presentation by Dr. Ben Caldwell (presented to another BSRB advisory committee in October) together at a subsequent meeting before drafting formal recommendations to the board. David Fye said staff will make that presentation available for committee viewing and follow-up discussion.