Artificial intelligence is no longer a future issue for addiction counselors. It is already entering our profession.
AI can summarize records, draft progress notes, assist with treatment plans, prepare patient education, analyze outcomes, support scheduling, transcribe sessions, identify patterns in large datasets, and communicate directly with consumers through chatbots and other digital tools.
Some of these applications could substantially reduce administrative burdens on an addiction workforce that desperately needs relief.
But we need to ask a different question before we ask what AI can do:
What should AI be allowed to do in addiction treatment?
That distinction may become one of the defining professional issues of the next decade.
For counselors, treatment providers, regulators, educators, credentialing organizations, and professional associations, AI presents enormous opportunities—but it also introduces significant questions involving confidentiality, professional responsibility, informed consent, clinical judgment, discrimination, documentation, liability, and the therapeutic relationship.
And the law is beginning to catch up.
Addiction Treatment Has a Different Privacy Problem
The addiction profession cannot simply adopt AI practices developed for general businesses.
Substance use disorder information can be subject to some of the strongest confidentiality protections in American health care.
In addition to HIPAA, federally assisted SUD programs may be subject to 42 CFR Part 2, the federal confidentiality regulations specifically protecting SUD patient records.
The federal government significantly revised Part 2 in 2024. Compliance with the revised requirements became mandatory on February 16, 2026. Among other changes, the rule more closely aligns certain Part 2 requirements with HIPAA, establishes breach-notification requirements, creates additional patient rights, and applies civil and criminal enforcement mechanisms.
This is not merely a paperwork change.
In February 2026, the U.S. Department of Health and Human Services Office for Civil Rights announced the beginning of its Part 2 civil enforcement program. OCR can now investigate complaints and use mechanisms that include resolution agreements, monetary settlements, corrective-action commitments, and civil money penalties.
That makes AI governance a compliance issue for the addiction field.
Imagine a counselor recording a counseling session using an AI transcription application.
- Where does that recording go?
- Is it stored?
- For how long?
- Who can access it?
- Is the vendor permitted to use the information to improve its models?
- What happens to the transcript?
- What contractual protections exist?
- Does the information include Part 2 records or HIPAA-protected health information?
- What happens if the AI company experiences a breach?
These are questions counselors and organizations should answer before using a product with patient information—not after.
The Counselor Still Owns the Professional Decision
There is another fundamental issue: responsibility.
Suppose AI generates a treatment-plan recommendation that is inappropriate.
Suppose it misses an important withdrawal risk.
Suppose an AI-generated progress note contains something the client never said.
Suppose an algorithm recommends a lower level of care.
Who is responsible?
The answer cannot simply be: “The computer recommended it.”
Professional accountability does not disappear because technology participated in the decision.
That principle is beginning to appear explicitly in state law.
California: Tell Patients When AI Is Talking to Them
California provides one example.
Under AB 3030, specified health facilities, clinics, physician offices, and group practices using generative AI to generate patient communications concerning clinical information generally must include a disclosure that the communication was generated by AI and provide instructions for contacting a human.
Importantly, the law provides an exception when the AI-generated communication has been read and reviewed by a licensed or certified health care provider.
That represents an important regulatory principle:
Human review matters.
California has also addressed AI and algorithms in utilization management through SB 1120, reflecting a broader concern about technology replacing individualized professional decision-making in health care.
This matters greatly to addiction treatment.
Anyone who has fought for authorization for residential treatment, withdrawal management, medication services, or continued care understands what can happen when complex human circumstances are reduced to boxes on a utilization-review screen.
Adding AI to that process could improve efficiency.
It could also automate bad decisions faster.
Illinois Has Gone Much Further
Illinois offers an even more direct warning for behavioral health professionals.
The Wellness and Oversight for Psychological Resources Act, Public Act 104-0054, became effective August 1, 2025.
Its stated purpose includes protecting consumers seeking therapy and psychotherapy from unqualified providers, including unregulated AI systems.
Illinois allows licensed professionals to use AI for specified administrative and supplementary support when the professional retains responsibility for interactions, outputs, and data use.
But there are boundaries.
The law prohibits using AI to make independent therapeutic decisions, directly engage clients in therapeutic communication, or generate therapeutic recommendations or treatment plans without professional review and approval.
The law also addresses something particularly relevant to counselors: AI recording and transcription of therapy sessions.
When permitted supplementary AI support involves recording or transcribing a therapeutic session, the professional must provide specified written information about the use and purpose of the AI and obtain consent.
Whether other states ultimately follow Illinois remains to be seen.
But counselors should pay attention.
Utah Is Regulating Mental Health Chatbots
Utah provides another approach.
State law requires a mental health chatbot to clearly and conspicuously tell a Utah user that it is artificial intelligence and not a human.
The disclosure must occur before access to the chatbot’s features and in additional circumstances specified by the statute.
This raises a much larger question for our profession.
When does a recovery chatbot become treatment?
Imagine an AI application that tells someone:
“It sounds like you’re experiencing cravings. Let’s work through them.”
Then:
“Based on what you’ve told me, I think you should try this coping strategy.”
Then:
“Tell me what happened in your childhood that you believe contributes to your substance use.”
At what point has technology moved beyond education and information into counseling?
That line will become increasingly important.
AI Is Also Coming From the Payer Side
Providers are not the only organizations adopting advanced technology.
CMS’s WISeR Model began in 2026 and tests technology-enhanced approaches to prior authorization and prepayment medical review for selected services in Original Medicare. CMS states that the model is intended to test whether these approaches can reduce waste while maintaining access to medically necessary care.
Although WISeR is not an SUD-specific program, the direction of travel matters.
AI will not only be something counselors and treatment programs use.
AI may increasingly be something used to evaluate providers, claims, documentation, medical necessity, utilization, and payment.
Our field therefore needs AI literacy on both sides of the equation.
We Need an AI Risk Continuum
One mistake would be treating every use of AI as equally dangerous.
Using AI to help develop a staff-meeting agenda is not the same as allowing AI to determine whether a patient requires residential treatment.
We need to evaluate AI according to risk.
Lower-risk applications might include:
Administrative scheduling, generic educational materials, internal brainstorming, non-patient-specific training materials, workflow assistance, or analysis of appropriately de-identified information.
Higher-risk applications include:
Processing identifiable patient information, generating clinical documentation, transcribing counseling sessions, communicating individualized clinical information, assisting with diagnosis, recommending treatment, assessing risk, recommending levels of care, or influencing admission and discharge decisions.
And then there should be a category that receives the greatest scrutiny:
AI acting independently in place of a qualified professional.
The closer AI gets to making consequential decisions about a patient’s treatment, safety, liberty, medication, level of care, or access to services, the greater the need for human accountability.
Ten Questions Counselors Should Ask Before Using AI
Before using an AI product professionally, counselors and treatment organizations should be able to answer these questions:
- What exactly is the AI being asked to do?
- Will patient-identifiable information be entered into the system?
- Could the information be protected by HIPAA or 42 CFR Part 2?
- Where is the information stored, and for how long?
- Can the company use our information to train or improve its AI?
- What happens if the vendor experiences a security breach?
- Does the patient need to be informed or provide consent?
- Is AI communicating directly with the patient?
- Is a qualified human reviewing consequential clinical outputs?
- Can we reconstruct what happened if an AI-assisted decision is later challenged?
If an organization cannot answer those questions, it probably is not ready to put patient information into the system.
Documentation Presents a Special Danger
Generative AI can write extraordinarily convincing clinical prose.
That is precisely why counselors should be careful.
A professionally written paragraph is not necessarily an accurate paragraph.
AI systems can generate incorrect information. In clinical documentation, even a small invented detail can have significant consequences.
Imagine an AI-generated note stating that a client denied suicidal ideation when the counselor never conducted that assessment.
Or documenting a specific intervention that never occurred.
Or changing the meaning of something a patient said.
Once the counselor signs the note, the issue is no longer simply whether the AI made a mistake.
The counselor has potentially authenticated inaccurate clinical documentation.
AI-assisted documentation therefore needs human review.
Not a quick glance.
Actual review.
The counselor must always maintain creative control.
We Cannot Automate the Therapeutic Relationship
There is something else at stake that cannot be measured by efficiency.
Addiction counseling is built around human relationships.
People frequently enter treatment carrying shame, trauma, distrust, stigma, broken relationships, and years of being reduced to diagnoses or behaviors.
Recovery often begins when another human being sees something in that person that the person cannot yet see in themselves.
Technology can support that relationship.
It should not quietly replace it.
A chatbot may be available at 2:00 a.m.
An algorithm may identify patterns a counselor missed.
AI may eventually help us personalize treatment in ways we cannot currently imagine.
We should welcome those possibilities.
But empathy is not merely producing empathetic words.
Counseling involves accountability, judgment, professional boundaries, cultural understanding, observation, experience, ethical responsibility, and a relationship between two human beings.
We should be extremely careful about confusing a machine’s ability to simulate therapeutic language with the professional practice of therapy.
Our Ethics Codes Will Need to Evolve
Professional associations and credentialing organizations also have work to do.
AI should become part of counselor education, continuing education, supervision, ethics training, and organizational compliance.
Future professional standards will increasingly need to address issues such as:
- AI-assisted documentation;
- patient confidentiality;
- informed consent;
- session transcription;
- clinical decision support;
- algorithmic bias;
- patient-facing chatbots;
- professional accountability;
- vendor selection;
- cybersecurity;
- documentation of AI use; and
- appropriate human oversight.
We should not wait for disciplinary cases to tell counselors what the boundaries are.
The Addiction Profession Should Help Write the Rules
Perhaps most importantly, addiction professionals need a seat at the table as policymakers begin regulating AI in behavioral health.
Technology companies will be there.
Health plans will be there.
Hospitals will be there.
Attorneys and regulators will be there.
The addiction treatment and recovery field must be there too.
SUD treatment has unique confidentiality requirements, workforce structures, peer services, residential settings, recovery support systems, reimbursement challenges, and therapeutic relationships that policymakers may not fully understand.
Rules written for hospitals or general medicine will not always translate neatly into community addiction treatment.
Our field should help states and the federal government develop policies that preserve innovation while establishing some fundamental principles:
- Patients should know when they are interacting with AI in circumstances where disclosure is appropriate or legally required.
- Patients should have meaningful access to human professionals.
- Protected SUD information should not become training data, advertising data, or a commercial commodity merely because it passed through an AI system.
- AI should not independently make high-consequence clinical decisions that require professional judgment.
- Organizations should know what their AI vendors do with patient information.
- Professionals should remain accountable for the clinical decisions they make with AI assistance.
- And patients should never have to surrender dignity, confidentiality, or meaningful human care simply because automation is cheaper.
The Future Should Be AI-Assisted, Not AI-Abandoned
I am optimistic about artificial intelligence.
I believe it could give counselors back time.
It could reduce administrative burden.
It could help small treatment programs perform sophisticated data analysis.
It could expand education.
It could help identify patterns that humans overlook.
It could strengthen quality improvement and help us understand outcomes.
And, used responsibly, it may allow addiction professionals to spend less time serving paperwork and more time serving people.
That is a future worth pursuing.
But innovation cannot outrun ethics.
The goal should never be to determine how much of addiction treatment we can remove the human being from.
The goal should be to determine how technology can help human professionals better support other humans in their recovery.
The organizations and professionals that succeed in the AI era will not necessarily be those that automate the most. They will be those that understand what to automate, what to regulate, what to disclose, what to protect—and what must always remain human.
References and Further Reading
U.S. Department of Health and Human Services, Office for Civil Rights. Fact Sheet: 42 CFR Part 2 Final Rule. Updated January 30, 2026. The revised Part 2 requirements became mandatory February 16, 2026.
U.S. Department of Health and Human Services, Office for Civil Rights. Understanding Confidentiality of Substance Use Disorder Patient Records or “Part 2.” Includes current information concerning patient notices, complaints, breach reporting, and the 2024 Final Rule.
U.S. Department of Health and Human Services. Office for Civil Rights Announces Civil Enforcement Program for Confidentiality of Substance Use Disorder Patient Records. February 13, 2026.
California Legislature. AB 3030 (2024), Health Care Services: Artificial Intelligence. Requirements concerning certain AI-generated patient clinical communications.
Illinois General Assembly. Public Act 104-0054, Wellness and Oversight for Psychological Resources Act. Effective August 1, 2025.
Illinois General Assembly. 225 ILCS 155/15. Requirements governing permitted professional use of AI, including certain recording and transcription uses.
Utah Legislature. Utah Code §13-72a-203. Disclosure requirements for mental health chatbots.
Centers for Medicare & Medicaid Services. WISeR Model Provider and Supplier Operational Guide and model materials. WISeR began implementation in 2026 for selected Original Medicare services.
This article is intended for professional education and discussion and does not constitute legal advice. Organizations should evaluate federal and state requirements applicable to their specific services, workforce, technology, and jurisdiction.