Conversational AI Safety Act [What It Changes]

TL;DR

• Conversational AI Safety Act is the emerging label for state laws that stop a public chatbot from advertising itself as a licensed mental-health professional, and, in some states, from delivering therapy at all.

• Idaho and Nebraska used that exact title, according to Holland & Knight's May 26, 2026 alert; both acts take effect July 1, 2027.

• Tennessee, Colorado, and Maine already flipped the switch in summer 2026: no "AI therapist" claim, and in Colorado a licensed clinician cannot let a bot run unsupervised therapy.

• A 2-truck HVAC shop, a 10-person agency writing ads for a wellness app, or a solo clinic with an intake bot should strip any "talk to our therapist" copy now, before the next state copies the same consumer-protection hook.

Key Takeaways

• The minted term is a statute title in two states and a policy family in several more. The shared rule is: a chatbot is not a licensee.

• Tennessee's SB 1580 makes an "AI therapist" ad an unfair or deceptive act, with a $5,000 civil penalty per violation, effective July 1, 2026.

• Colorado's HB26-1195 (signed June 3, 2026) goes further: it limits how psychologists, counselors, social workers, and marriage and family therapists may use AI with clients, and it treats fake-license marketing as a Colorado Consumer Protection Act violation.

• According to the Transparency Coalition's July 27, 2026 roundup, 5 states prohibited offering AI chatbot-based therapy.

• Self-help, journaling, meditation, and FDA-regulated tools get Colorado carve-outs. "I am your licensed counselor" does not. Congress has not passed a matching federal healthcare-AI statute.

What Conversational AI Safety Act means

Conversational AI Safety Act is the name Idaho and Nebraska gave to statutes that require a public-facing chatbot to disclose that the user is talking to a machine, to route suicidal ideation to a crisis protocol, and to stop the bot from claiming it provides professional mental or behavioral healthcare.

Tennessee, Colorado, Maine, Vermont, and Rhode Island wrote the same policy in other titles. Wellness apps, telehealth intake bots, and agency landing pages that still say "chat with a therapist 24/7" are the operational target. A 10-person agency is in this statute family the moment an ad implies a license. A solo clinic that lets an intake bot "start the session" before a human joins is in it under Colorado's synchronous-interaction rule.

Teams that already route intake messages through US Tech Automations workflows can add a claim-review step on outbound chatbot copy rather than rebuild the rest of the stack. US Tech Automations is the site behind this explainer.

Who shipped the statutes, and when

Holland & Knight, writing on May 26, 2026, is the source that records the title. According to that alert, Idaho (March 1, 2026) and Nebraska (April 8, 2026) enacted nearly identical laws, both titled the Conversational AI Safety Act, effective July 1, 2027. The acts require operators of public-facing AI chatbots to disclose the AI, mandate crisis-response protocols for suicidal ideation, and prohibit chatbots from representing that they provide professional mental or behavioral healthcare.

Tennessee moved earlier on the advertising ban. The enacted Senate Bill 1580 text adds Tenn. Code Ann. § 33-1-205: a person who develops or deploys an artificial intelligence system shall not advertise or represent to the public that such system is or is able to act as a qualified mental health professional. A violation is an unfair or deceptive act under the Tennessee Consumer Protection Act of 1977, with a civil penalty of five thousand dollars ($5,000) per violation. Section 3 of the act takes effect July 1, 2026.

Colorado's companion is HB26-1195, Psychotherapy Artificial Intelligence Restrictions. Status is "Became Law." According to the Colorado General Assembly bill page, the governor signed on June 3, 2026. The enacted summary prohibits regulated professionals from allowing an AI system to interact with clients in therapeutic communication without synchronous, real-time interaction among the professional, the AI system, and the client, and from generating therapeutic recommendations or treatment plans without review and approval. Using terms that imply the AI's outputs are equivalent to a regulated professional, or that the AI provides psychotherapy, is an unfair or deceptive trade practice under the Colorado Consumer Protection Act. Administrative support is allowed if the professional reviews outputs. Recording a session through AI requires advance disclosure and written informed consent.

The Transparency Coalition dated July 27, 2026 lists Colorado HB 1195 as enacted June 3, 2026 and effective August 12, 2026; Maine LD 2082 as enacted April 13, 2026 and effective July 29, 2026; Vermont H 816 as enacted and effective June 17, 2026; and Rhode Island H 7349 / S 2197 as enacted June 22, 2026 and effective January 1, 2027. Holland & Knight, writing before Colorado's signature, described Maine's HB 2082 (April 8, 2026) as allowing licensed professionals to use AI only for administrative and limited supplementary purposes, and barring AI for therapeutic communications, treatment decisions, or independent patient interaction.

Oregon took the companion-bot path. SB 1546 is enrolled as Chapter 85 and requires operators of AI companions to notice users that they are interacting with artificial output if a reasonable person would believe they were interacting with a natural person. Holland & Knight adds crisis referrals, minor safeguards, incident reporting, and a private right of action, effective January 1, 2027.

Why now

The constraint that broke is licensure. NAMI's overview of mental-health professionals lists psychologist, counselor, social worker, and marriage and family therapist as state-licensed roles. A chatbot does not hold a license. Holland & Knight is explicit that the U.S. Congress has yet to pass significant legislation regulating AI in healthcare. States filled the gap.

The Transparency Coalition's 2025 report counted 73 new AI-related laws in 27 states last year, including an Illinois ban on commercial AI therapy chatbots. The 2026 mid-year report, released July 21, 2026, counted 84 new AI-related laws across 27 states so far in 2026, already past last year's full-year total.

Crisis traffic is the political fuel. SAMHSA's 988 page states that in 2020 the U.S. had one death by suicide every 11 minutes, and that suicide is a leading cause of death for people aged 10–34. According to CDC FastStats, U.S. suicide deaths numbered 48,824, or 14.4 per 100,000 population, ranked 10th among causes of death (2024 mortality via CDC WONDER). Idaho, Nebraska, and Oregon wrote suicidal-ideation protocols into chatbot law. The 988 Lifeline is the human number those statutes point at.

The 2026 chatbot and therapy table

Sources: Holland & Knight; Transparency Coalition July 27; Colorado HB26-1195; Tennessee SB 1580; Oregon SB 1546.

Parallel track: insurers may not let a model deny care alone

The same session class also hit prior authorization. A clinic that sells both a chatbot and a payer contract will meet both.

Sources: Holland & Knight; Transparency Coalition; Colorado HB26-1139.

Colorado's insurer bill HB26-1139 was signed June 2, 2026, effective January 1, 2027 (chapter 325), and tells carriers and Medicaid payers they may not pay for psychotherapy conducted by an AI system. The 2024 Medicare Advantage and Part D final rule (CMS-4201-F) already requires a minimum 90-day transition when an enrollee on treatment switches MA plans, plus an annual Utilization Management Committee review. Holland & Knight notes CMS has said MA plans may use AI to assist prior authorization only if the tools account for the beneficiary's unique clinical conditions.

What a small shop must actually change

If you run a clinic intake bot, a wellness app, or a studio membership chat, three copy tests now exist. First: does any string say the system is a therapist, counselor, psychologist, social worker, or qualified mental health professional? Tennessee forbids that representation. Colorado forbids implying that outputs are equivalent to a regulated professional. Idaho and Nebraska forbid representing that the chatbot provides professional mental or behavioral healthcare.

Second: does a licensed human stay in the therapeutic loop in real time? Colorado requires synchronous interaction among the professional, the AI system, and the client. Maine, per Holland & Knight, bars independent patient interaction by AI. Third: if a session is recorded or transcribed by AI, did the client get advance disclosure and written consent? Colorado says yes.

Carve-outs exist. Colorado does not apply the ban to technology that does not diagnose or treat mental-health disorders, clearly discloses it is not a substitute for clinical care, and provides self-help, homework, coaching, navigation, meditation, journaling, or is FDA-regulated. The FDA's device-software page is the overlay for apps that meet the device definition in section 201(h) of the FD&C Act.

Read patient-engagement tools against this list: the patient-engagement roundup, the healthcare automation survey, and Epic vs athenahealth. A boutique studio is not a clinic, but a "wellness coach bot" that slides into therapist language is; see Mindbody vs WellnessLiving, one-location studio automation, and independent trainer tools.

A clinic that already uses US Tech Automations to file patient-engagement messages can park any "start therapy" prompt behind a licensed-user gate and a disclosure string.

Federal floor that still applies

State chatbot statutes sit on older federal rails.

The FTC Act (15 U.S.C. §§ 41–58) empowers the Commission to prevent unfair or deceptive acts or practices. FTC truth-in-advertising guidance says an ad must be truthful, not misleading, and, when appropriate, backed by scientific evidence, with extra scrutiny for health claims. Tennessee and Colorado wired their AI-therapist bans into those same buckets.

Wellness apps that are not HIPAA covered entities still meet the FTC Health Breach Notification Rule (16 CFR Part 318) when they hold personal health records; breaches affecting 500 or more people also require media notice. Covered clinics still have the HIPAA Privacy Rule at 45 CFR Parts 160 and 164. Colorado treats a chatbot claim that user data is "confidential like therapy" as deceptive if a reasonable user would think therapist-client privilege applies. Do not let marketing copy invent a privilege the bot does not have.

The APA Ethics Code (effective June 1, 2003, amendments 2010 and January 1, 2017) still binds psychologists, including Standard 3.04 on avoiding harm. NIST's AI Risk Management Framework (AI RMF 1.0, January 26, 2023) is voluntary and does not license chatbots; a generative-AI profile (NIST.AI.600-1) followed on July 26, 2024. Holland & Knight's Health AI Navigator is a pointer, not the statute.

USTA analysis: the 42-day and 365-day clocks

This is a USTA analysis. Inputs are the effective dates already cited: Tennessee July 1, 2026 (statute text); Colorado August 12, 2026 (Transparency Coalition); Idaho and Nebraska July 1, 2027 (Holland & Knight).

Sources: Tennessee SB 1580; Transparency Coalition; Holland & Knight; Colorado HB26-1195 vote tables; TCAI mid-year. USTA analysis = arithmetic on those inputs only.

Read the 42 days as the window a multi-state app had between Tennessee's advertising ban and Colorado's practice ban. Read the 365 days as the lag before the two states that actually named the Conversational AI Safety Act turn their title on. A shop that only watches the named title will be a year late.

Honest limits

These are state statutes. They do not create a national license for chatbots. Holland & Knight (May 26) and the Transparency Coalition (July 27) disagree on some enacted dates by a few days (Tennessee April 6 vs April 1; Maine April 8 vs April 13). Where the enrolled text is public, use the text. Colorado's official page confirms signature on June 3, 2026; the August 12, 2026 effective date is the Transparency Coalition's figure. Idaho's and Nebraska's enrolled texts were not retrieved on this pass; title, duties, and July 1, 2027 come from Holland & Knight only. This is not legal advice.

Signal vs Speculation

Signal (sourced). Idaho and Nebraska titled statutes the Conversational AI Safety Act, effective July 1, 2027. Tennessee SB 1580 is in force as of July 1, 2026, with a $5,000 per-violation civil penalty. Colorado HB26-1195 is law as of the June 3, 2026 signature. Maine, Vermont, and Rhode Island passed 2026 therapy-chatbot measures. Oregon SB 1546 is Chapter 85 on AI companions. Five 2026 states appear on the Transparency Coalition's therapy-chatbot-ban list. 84 AI-related state laws passed in 27 states in 2026 H1. Congress has not passed a matching healthcare-AI statute. FTC, HIPAA, FDA software policy, APA ethics, and CMS MA prior-auth rules still apply.

Our read (12–36 months, small and mid-size clinics and wellness shops). If the $5,000 Tennessee penalty is assessed on a public app, copy-cat advertising bans will move faster than full practice bans, because an unfair-practice hook does not need a new license board. If assessments stay at zero, agencies will keep shipping "AI therapist" headlines until a state AG picks a defendant. Colorado's synchronous-interaction rule is the operational one for a licensed practice: a bot that chats while the clinician is in another room is the fact pattern that statute stops. Studio apps that stay inside journaling, scheduling, and coaching, and that disclose they are not clinical care, still fit Colorado's carve-out. We do not forecast a federal Conversational AI Safety Act. We do expect more states to reuse the title Idaho and Nebraska already printed.

Frequently asked questions

Does the Conversational AI Safety Act ban all healthcare chatbots?

No. Idaho and Nebraska, as described by Holland & Knight, target public-facing chatbots that fail to disclose AI, fail a crisis protocol, or claim to provide professional mental or behavioral healthcare. Scheduling and billing bots that do not make that claim are a different product.

What does Tennessee actually prohibit?

The SB 1580 text prohibits a developer or deployer from advertising or representing to the public that an AI system is or is able to act as a qualified mental health professional.

Can a licensed Colorado psychologist use AI at all?

Yes, with limits. HB26-1195 allows administrative and supplementary support if the professional reviews outputs, plus educational, simulation, training, and IRB-overseen research. Therapeutic communication without a synchronous human is out.

When did Colorado's psychotherapy AI act take effect?

The Transparency Coalition lists August 12, 2026. The official bill page lists governor signature on June 3, 2026 and status "Became Law."

Do HIPAA and the FTC still apply if my state has no chatbot statute?

Yes. Covered entities still follow the HIPAA Privacy Rule. Health apps that hold personal health records still meet the FTC Health Breach Notification Rule. Ads still must be truthful under the FTC Act.

What should a wellness studio put in the bot's first message?

A plain statement that the user is talking to software, that it is not a licensed clinician, and that someone in crisis should call or text 988, matching the 988 Lifeline.

Glossary

• Conversational AI Safety Act: The title Idaho and Nebraska used for 2026 chatbot-safety statutes; here also the label for the wider 2026 family of AI-therapist advertising and practice bans.

• Qualified mental health professional: The phrase Tennessee's § 33-1-205 forbids an AI system from claiming.

• Regulated professional (Colorado): A person lawfully permitted to provide psychotherapy in the state.

• Synchronous, real-time interaction: Colorado's required three-way loop among clinician, AI system, and client.

• Unfair or deceptive act: The consumer-protection hook Tennessee and Colorado use instead of a new AI agency.

• AI companion: Oregon SB 1546's object: a system a reasonable person might think is a natural person.

• 988: The national suicide and crisis lifeline operated with SAMHSA, available 24/7.

What to do next

Open your chatbot strings. Delete any claim that the system is a therapist, counselor, or qualified mental health professional. Add a 988 line. If a licensed clinician uses the bot with a patient in Colorado or Maine, keep that clinician in the live loop.

If those checks already sit in a workflow graph, add them as a gate, not as a rebuild. A clinic that files engagement messages through US Tech Automations can put the disclosure and license-claim scan on the same path as the send step. Open the agentic workflow builder and attach that gate to the chatbot node.