Artificial intelligence documentation tools have moved from novelty to mainstream in clinical practice remarkably quickly. Therapists in private practice, community mental health, and hospital settings are now using AI to draft progress notes, generate treatment plan language, and summarize session content. The efficiency gains are real — clinicians report saving 30 minutes to two hours per day on administrative tasks. But efficiency alone does not make AI use ethical. Several distinct ethical obligations apply when you introduce AI into clinical documentation.
Should Clients Know You Are Using AI?
This question sits at the intersection of informed consent, professional transparency, and client autonomy. Most professional ethics codes do not yet contain explicit AI-specific provisions, but general principles apply clearly. The APA Ethics Code requires informed consent that covers procedures used in treatment. The NASW Code of Ethics emphasizes transparency about tools and methods. The AAMFT Code similarly stresses client autonomy and informed participation.
Practically, if an AI tool is processing session content — even in summarized form — clients have a legitimate interest in knowing this. A reasonable informed consent disclosure should explain: (1) that you use an AI tool to assist with note documentation, (2) what data the tool processes and how, (3) where that data is stored and for how long, and (4) that you, the clinician, review and are responsible for all final documentation. Many clinicians add this language to their standard informed consent forms. Some state licensing boards are beginning to issue guidance specifically requiring disclosure; check your board's current position.
The Clinician Remains Responsible for Every Word
AI documentation tools generate drafts. They do not generate records. This distinction is legally and ethically critical. The clinician who signs a note is the author of that note, regardless of whether AI generated the initial language. If an AI tool drafts a progress note that inaccurately characterizes a client's risk level, and the clinician signs it without correction, the clinician bears full professional and legal responsibility for that inaccuracy.
This means AI-generated drafts must be read carefully, edited for accuracy, and revised to reflect the clinician's actual clinical judgment. Common AI errors include: paraphrasing session content inaccurately, missing nuance in risk assessment language, generating generic language that doesn't match the specific client, and occasionally confabulating details. Reviewing AI output requires the same critical attention you would give to a trainee's note before countersigning — because the liability is identical.
Scope of AI Assistance: What AI Can and Cannot Do
AI tools can legitimately assist with: structuring notes into a consistent format, generating draft language based on prompts you provide, identifying documentation elements you may have missed (like missing a treatment goal update), and reducing the cognitive load of translating session events into written language.
AI tools cannot appropriately: make clinical diagnoses, determine risk level independently, decide on level of care, formulate clinical conceptualizations, or replace clinical judgment in any area that requires professional training. Any AI output that presents itself as a clinical conclusion rather than a draft for your review should be treated with particular skepticism. When using AI tools, maintain a clear internal distinction between AI-assisted formatting and AI-generated clinical reasoning — the latter is never acceptable.
Data Privacy: Where Are Your Notes Being Processed?
This is the technical question clinicians most commonly underestimate. When you use an AI documentation tool, your session content — which constitutes protected health information (PHI) — is transmitted to and processed by the tool's servers. The critical questions are:
Where are those servers located, and who can access the data? Is the data used to train the AI model? If so, you are contributing client PHI to a commercial training dataset, which creates serious HIPAA and ethical problems. What is the data retention period? Is the data encrypted in transit and at rest using current standards (TLS 1.2 or higher, AES-256 encryption)?
Reputable AI documentation tools designed for healthcare will provide clear answers to all of these questions, execute a Business Associate Agreement (BAA), and explicitly state that they do not train on client data. Vendors who cannot or will not answer these questions clearly should not be used for clinical documentation.
HIPAA Compliance of AI Vendors
Under HIPAA, any vendor who handles PHI on your behalf is a Business Associate. Business Associates must sign a BAA with your practice before you share PHI with them. This applies to AI documentation tools without exception. The BAA contractually obligates the vendor to protect PHI, report breaches, and limit use of data to the purposes specified in the agreement.
Before using any AI tool with clinical content, verify: (a) the vendor will execute a BAA, (b) the BAA covers the specific use case (documentation assistance), and (c) the vendor has documented security practices that meet HIPAA Security Rule requirements. Vendors who offer "HIPAA-compliant" features only on enterprise tiers, but let you use the free tier with clinical content, are not providing HIPAA-compliant service at the tier you are using.
What Licensing Boards Are Saying
Licensing board guidance on AI documentation is actively evolving. As of 2025, several state psychology boards and counseling boards have issued position statements. Common themes include: AI tools do not change the clinician's responsibility for the accuracy of records; clients should be informed when AI is used in their care; clinicians must have sufficient understanding of AI tools to supervise their use; and AI-generated notes require review before signing.
Check your state licensing board's website for current guidance, and consult your professional liability insurance carrier for any coverage considerations specific to AI tool use. Some carriers now ask about AI documentation practices on renewal applications.
Liability When AI Output Contains Errors
The legal framework here is straightforward: when you sign a clinical note, you attest to its accuracy. If an AI tool generated a note saying a client denied suicidal ideation when in fact the client expressed passive ideation that you assessed as low risk, and you signed the note without correction, you have created a false record. The consequences range from licensing board complaints to malpractice liability depending on the clinical outcome.
Protect yourself by: always reading AI drafts before signing, editing language that doesn't precisely reflect the session, and documenting that clinical judgment was applied (not just that AI drafted the note). Some clinicians add a standard footer noting "This note was drafted with AI assistance and reviewed and approved by the clinician of record" — this makes explicit what the record represents.
Best Practices for Ethical AI Use in Documentation
Develop a consistent workflow: use AI to generate a draft, then spend five to ten minutes reviewing and editing before signing. Never sign a note you haven't read in full. Train any staff who use AI documentation tools on the same review requirements. Update your informed consent documents to disclose AI use. Verify BAAs are in place before using any new tool. Maintain documentation of your AI tool vendor's security and privacy practices in case of a licensing board inquiry.
Evaluating an AI Tool's Ethical Framework
Before adopting an AI documentation tool, ask the vendor directly: Do you train on client data? Who can access the data processed through your platform? What is your breach notification procedure and how fast will you notify me? Have you been independently audited for HIPAA compliance? Will you sign a BAA? A vendor committed to ethical AI use will answer these questions readily. A vendor that deflects, hedges, or buries the answers in fine print is telling you something important about their priorities.
AI tools can genuinely improve the sustainability of clinical practice by reducing documentation burden. But the ethical framework for their use is the clinician's responsibility to construct and maintain, not the vendor's.