Frequently Asked Questions: Ethical Use of Generative AI in ABA Practice

1. Is it ethical for BCBAs to use AI to draft session notes?

Using AI to create a first draft is not inherently unethical, but the clinician must review, correct, and sign the note as the responsible professional. Code 2.09 requires accurate documentation.

If AI-generated language misrepresents what occurred in a session—which is possible when the AI extrapolates beyond the data provided—the practitioner who signs the note is accountable for that inaccuracy. Davis et al.

(2026) found that training on new software requires addressing clinical judgment alongside procedural steps. The same principle applies: learning to use AI includes learning to verify its outputs.

2. Which BACB Ethics Code provisions are most relevant to AI use?

Code 2.01 (Providing Effective Treatment based on current scientific knowledge), Code 2.02 (Confidentiality), Code 2.09 (Accurate Documentation), and Code 1.01 (Maintaining Competence) are the four most directly applicable. Code 6.01, which addresses honest and accurate professional statements, also applies when AI-generated content is published or presented as clinical work.

Practitioners should review each provision against their specific AI use cases before adoption.

3. How do I evaluate whether an AI tool adequately protects client confidentiality?

Review the tool's data processing agreement for whether user inputs are retained, used for model training, or shared with third parties. For tools used with client data, a HIPAA-compliant business associate agreement is required.

If the tool cannot provide documentation of these protections, it should not receive client-identifiable information. Many consumer AI tools do not meet this threshold and should be restricted to non-client-specific tasks.

4. Can AI replace a functional behavior assessment?

No. A functional behavior assessment requires direct observation, data collection across environments, systematic manipulation of antecedents and consequences, and clinical interpretation by a credentialed practitioner.

AI can assist with organizing and summarizing data already collected, but it cannot conduct observation, identify motivating operations through direct assessment, or make individualized clinical judgments that Code 2.01 and Code 2.09 require. Using AI-generated FBA reports without independent assessment is a clinical and ethical failure.

5. What is the risk of AI generating inaccurate behavioral data?

Bartle et al. (2026) found that training exemplar quality directly affected practitioner procedural fidelity—a principle that applies to AI outputs.

AI generates plausible-sounding data summaries based on statistical patterns, not actual measurement. Frequency counts, percentage-correct values, or trend descriptions produced by AI can be internally consistent but factually incorrect.

Practitioners must always verify AI data summaries against original raw data before including them in any clinical record.

6. Do I need to disclose AI use to clients or caregivers?

The BACB Ethics Code does not yet contain an explicit AI disclosure requirement, but Code 2.01's transparency requirements and Code 2.02's confidentiality provisions together suggest that informed consent language addressing AI use is appropriate. If AI is used to process documentation or generate content, caregivers have a legitimate interest in understanding how their information is handled.

Practitioners who develop clear AI disclosure language now will be ahead of forthcoming formal guidance from the BACB.

7. What does augmentation vs. replacement mean in clinical practice?

Augmentation means AI assists a task that the clinician still owns and validates. The clinician prompts, reviews output, corrects inaccuracies, and accepts professional responsibility for the final product.

Replacement means AI output goes directly into clinical use without independent verification. The ethical risk concentrates in replacement uses.

Long et al. (2026) established that feedback quality in skill training requires qualified observation—a property AI lacks.

Practitioners cannot transfer professional accountability to a statistical model.

8. How should supervisors address AI use with supervisees?

Supervisors should develop explicit written policies before AI questions arise. Policies should specify which tools are approved for which tasks, what verification procedures are required before AI outputs are incorporated into documentation, and how AI use should be disclosed in clinical records.

Training should address not just how to use AI tools but how to evaluate their outputs critically—the same judgment standard required for any other indirect data source.

9. Are there AI applications in ABA that carry lower ethical risk?

Lower-risk applications typically involve AI operating on non-identifiable data or producing outputs clearly labeled as preliminary and requiring clinician review. Jiang & Wang (2026) documented that AI-assisted pattern analysis in AAC data showed clinical value when validated by qualified clinicians—that validation requirement is the distinguishing feature of lower-risk use.

Applications that directly contact clients, generate individualized clinical recommendations, or produce records signed without independent review carry substantially higher risk.

10. What should I do if my employer requires AI tools I believe raise ethical concerns?

Code 3.14 addresses conflicts between employer requirements and ethical obligations, establishing that behavior analysts must not comply with organizational demands that require violating the ethics code. If an employer requires use of an AI tool that cannot meet confidentiality or data integrity standards, the practitioner's obligation is to raise the concern formally, document it, and seek guidance from BACB ethics resources if the concern is not resolved internally.