AI tools are arriving in behavioral healthcare faster than the professional conversation is keeping up. BCBAs are right to ask hard questions: What does the ethics code say about this? Where is the line between a legitimate training tool and something that creates liability? And how do you evaluate a vendor’s claims before trusting their software with anything that touches your practice?
This post is a starting point for the ethical reasoning BCBAs should already be doing — because the tools are here whether the field is ready or not.
Two Very Different Use Cases
The first thing to clarify: “AI in behavioral healthcare” describes at least two fundamentally different things, and they carry different ethical weight.
AI for practitioner training and simulation places AI on the other side of a practice scenario. A BCBA or RBT practices a clinical interaction — delivering instructions, applying a prompt hierarchy, managing an escape-maintained behavior — with an AI persona that plays the client role. No real client is involved. The AI’s output is a learning experience for the clinician, not a treatment decision.
AI embedded in direct treatment uses algorithms to analyze session data, suggest programming adjustments, or interact with clients in any capacity. This is a different category entirely, with higher stakes and far less professional consensus.
These two use cases deserve separate ethical frameworks. Conflating them — treating all AI in behavioral healthcare as equivalent — leads to either premature rejection of useful training tools or uncritical adoption of tools that warrant much more scrutiny.
What the BACB Ethics Code Says (and What It Doesn’t)
The BACB’s Ethics Code for Behavior Analysts does not yet address AI simulation tools by name — the field is moving faster than any ethics code can. But the principles that govern professional conduct translate directly.
Competence. BCBAs are responsible for working within their areas of competence and for ensuring that the tools and procedures they use meet professional standards. If you adopt an AI training tool, you are responsible for understanding what it does, how it generates its outputs, and whether the simulated interactions it produces are clinically accurate. Outsourcing clinical judgment to a vendor’s interface is not a defense.
Supervisory responsibility. When you recommend or require AI simulation for supervisees, you carry ethical responsibility for that recommendation. The tool should build clinical skill, not model bad technique. Supervisors should review what their RBTs are practicing against — including whether the AI persona responds in ways that reflect realistic learner behavior.
Client welfare and informed consent. For simulation tools that are fully decoupled from client treatment — where no client data enters the system and no AI output influences a treatment decision — the consent considerations are minimal. The moment any client data is involved, informed consent and confidentiality protections apply fully, including HIPAA requirements for protected health information.
Do not use technology to circumvent ethics. BCBAs cannot use third parties — or third-party tools — to do what they themselves are prohibited from doing. If an AI tool makes a claim you would be uncomfortable making in your own clinical notes, that is a signal worth taking seriously.
Four Questions to Ask Before Adopting Any AI Tool
When evaluating an AI tool for your practice — simulation or otherwise — these four questions cut through the marketing:
-
Who has access to the data? Even for a simulation tool, understand what happens to session transcripts, performance records, and any identifying information you or your supervisees enter. Read the privacy policy, not just the sales deck.
-
What is the AI actually doing? Ask for a plain-language explanation of how the model generates its outputs. If the vendor cannot explain it without jargon, that is a problem — not a selling point.
-
Does it model clinically accurate behavior? For simulation tools, the AI persona should respond in ways that reflect real learner behavior patterns. Inaccurate simulation trains bad habits. Ask how the vendor validates clinical accuracy and who was involved in developing the scenario content.
-
Is the tool positioned as a training aid or a clinical decision-maker? These are not the same thing. Any vendor blurring this line deserves skepticism.
The Distinction That Matters Most
The most important ethical line in AI for behavioral healthcare is the one between building practitioner competence and replacing practitioner judgment.
Simulation tools that help BCBAs and RBTs accumulate deliberate practice reps — before they face a real client — support the ethics code’s requirements around competence and training. A BCBA who arrives at their first challenging client interaction having already navigated dozens of simulated scenarios is better prepared than one who hasn’t. That preparation is not a shortcut; it is the point of training.
AI that makes decisions about what a client should work on next, interprets behavior function without BCBA review, or generates treatment recommendations without a licensed clinician in the loop is a different matter — and one where the field should move carefully, with input from BACB and the broader behavior analytic community.
How Kipr Approaches This
Kipr is a simulation tool for practitioner training. AI personas play client roles; BCBAs and RBTs play themselves. No client data enters the system. The purpose is reps — clinical judgment built through practice before practitioners face real clients in high-stakes moments.
Every scenario Kipr generates is designed to be reviewed and contextualized by a supervising BCBA. The simulation is the practice environment. Clinical decisions stay with the clinician. That boundary is not incidental to what Kipr does — it is the product’s core design principle.
BCBAs who ask hard questions before adopting any AI tool are doing exactly what the ethics code requires. We think the same way.
Join the Kipr waitlist for early access.