Ethics of AI in ABA — Frequently Asked Questions

1. Does using an AI tool to draft session notes violate the BACB Ethics Code?

Not automatically, but it can. The ethical concern is not the use of the tool itself but whether the documentation it produces is clinically accurate, individualized, and signed off by a BCBA who has genuinely reviewed it.

If you use an AI draft as a starting point, review every sentence for accuracy, revise what is not accurate, and can attest that the final document reflects your professional judgment — that is defensible. If you publish AI-generated notes with minimal review, you are certifying clinical content you have not actually verified, which violates your competence and integrity obligations under the Ethics Code.

2. What should I disclose to families about AI use in their child's care?

The principle of respect for client autonomy, which runs throughout the BACB Ethics Code, suggests that families should know in what ways AI tools are being used in the services they receive. At minimum, disclosure should cover whether AI tools contribute to treatment plan development, session documentation, or caregiver communications.

Families should have the opportunity to ask questions about how those tools work and to express preferences about AI involvement in their care. Some families will be comfortable with AI assistance and some will not; both responses deserve to be honored within the constraints of your practice setting.

Document your disclosure conversations in the case file.

3. How do I evaluate whether an AI tool is appropriate for clinical use?

A structured evaluation process should address four questions: What specific function does the tool perform, and does that function touch clinical judgment? What validation evidence exists for the tool's accuracy and reliability in populations like your clients?

What are the known failure modes, and how does the system fail when it fails? Who bears accountability when the tool makes an error?

AI tools that cannot produce transparency on these questions should be used — if at all — only for low-stakes administrative functions where errors carry minimal clinical consequence. For any tool that informs clinical decisions, validation data specific to diverse client populations should be a prerequisite, not a nice-to-have.

4. What is the relationship between AI-assisted authorization decisions by insurers and my clinical obligations?

When a payer's algorithmic system approves or denies authorization for a client's treatment, the BCBA's ethical obligations do not change. If the authorized hours are clinically insufficient for the client's needs, the BCBA is obligated to advocate for the client, document the clinical rationale for additional hours, and pursue appeals where appropriate.

The fact that the authorization decision was generated by an algorithm rather than a human reviewer does not transfer clinical judgment to the algorithm. The BCBA remains responsible for ensuring the client receives a level of service that meets their assessed needs, and documentation of the advocacy process is essential protection if outcomes are later scrutinized.

5. Can AI tools assist with functional behavior assessment, or is that outside current AI capabilities?

Current AI tools cannot conduct functional behavior assessments. Functional assessment requires direct observation, structured interviews conducted with clinical judgment about follow-up questions, and interpretation of behavioral data in the context of the individual learner's history, environment, and reinforcement landscape.

What AI tools can do is help organize and summarize data that has been collected by human observers — identifying patterns in structured data, flagging sessions that deviate from trend, or generating summary language from clinician-entered data. These are genuinely useful applications, but they are assistants to functional assessment, not substitutes for it.

6. How should practice owners set up supervision systems when AI tools are being used by frontline staff?

Supervisory systems must explicitly account for AI-assisted workflows, or they will fail to catch the specific types of errors AI tools introduce. Add an AI documentation audit to your regular supervision checklist: are AI-generated notes individualized and clinically accurate, or are they templated and generic?

Are treatment goals copied from AI-generated language, or are they grounded in current functional assessment? Spot-check a percentage of AI-assisted documents each week and give direct feedback on any AI-specific error patterns.

Ensure that BCBAs under your supervision can explain the clinical basis of every statement in their documentation — if they cannot, the AI tool has displaced clinical judgment rather than supporting it.

7. What are the specific risks when AI is used to generate caregiver-facing materials?

Caregiver-facing materials generated by AI carry several specific risks. First, they may use language calibrated for a general population that does not match the reading level, cultural context, or specific concerns of the family receiving it.

Second, they may include clinical recommendations that are accurate as general principles but incorrect for the specific learner — a child whose reinforcers or behavioral function differs from the cases the AI was trained on. Third, they can create an impression of thorough, individualized guidance when the underlying content is boilerplate.

Families who follow AI-generated caregiver training materials that are misaligned with their child's actual needs may implement procedures incorrectly, potentially strengthening maladaptive behaviors rather than reducing them.

8. What does social validity mean in the context of AI-assisted ABA practice?

Social validity in ABA refers to the social acceptability of the goals, procedures, and outcomes of an intervention from the perspective of the client, family, and broader community. AI-assisted practice raises social validity concerns at multiple levels.

Are families consulted about whether they are comfortable with AI involvement in their child's care? Do AI-generated treatment goals reflect the outcomes the family actually prioritizes, or the outcomes that most commonly appear in training data?

Do the procedures recommended by AI tools align with community standards for what constitutes compassionate, ethical care? BCBAs must actively solicit feedback on these dimensions rather than assuming that efficient AI-assisted documentation translates into socially valid services.

9. How should I think about intellectual property and data privacy when using AI tools with client data?

Every AI tool that processes client data raises HIPAA and data privacy considerations that your practice must address before adoption. Review the vendor's data processing agreements carefully: does the tool store client data on its servers, use client data to train future models, or share data with third parties?

Ensure that any AI platform you use has a signed Business Associate Agreement before processing any protected health information through it. Beyond legal compliance, consider the ethical dimension: clients and families who share sensitive behavioral and developmental information are trusting that information to your professional care.

Using it to train commercial AI systems without explicit informed consent raises legitimate autonomy concerns that the field has not yet fully resolved.

10. Are there AI applications in ABA that are clearly beneficial and low-risk?

Yes. AI tools that operate at the level of data organization and administrative efficiency, without generating clinical recommendations or documentation content, carry substantially lower risk.

Automated scheduling systems, data graphing tools that visualize behavior data without interpreting it, search and retrieval systems that help BCBAs locate relevant research or training resources, and tools that help organize billing and authorization documentation are examples of relatively low-risk applications. The risk level rises as the AI function moves closer to clinical judgment — treatment recommendations, goal language, functional assessment interpretation — and practice owners should assign oversight requirements that scale accordingly.