.jpg)
On 14 April 2026, Governor Jim Pillen of Nebraska signed into law the Conversational Artificial Intelligence Safety Act, contained within LB 525, which the state's unicameral legislature had approved unanimously four days earlier by 49 votes-to-none. The Act took effect on 17 April. Four days before the Nebraska signing, Maine had sent its own Legislative Document 2082 to the governor's desk, a narrower measure that prohibits unlicensed individuals, including artificial intelligence systems, from providing therapy or psychotherapy.
The two bills have often been reported in the same breath, as parts of a single regulatory wave. They are worth reading separately. They do different things, through different instruments, and together they begin to describe the shape that American conversational AI law is now taking.
What Maine did
Maine's LD 2082 is a licensing move. It extends an existing professional rule, that therapy and psychotherapy require a licensed practitioner, to cover delivery through AI systems. Clinical functions are prohibited to unlicensed providers. Administrative functions, scheduling, record keeping, and intake support, are permitted. The bill draws a bright line at the practice of therapy itself, using the enforcement architecture that already governs unlicensed practice.
What Nebraska did
Nebraska's Act is a different instrument. It does not restrict itself to therapy or to mental health practitioners. It regulates conversational AI as a class. It requires an operator to disclose to the user that the service is not human, where a reasonable person would otherwise be misled. It requires the adoption of a protocol for responses to expressions of suicidal ideation or self-harm, including a referral to crisis services. And it prohibits the service from representing, explicitly, that it is designed to provide professional mental or behavioural health care.
The Act also contains targeted obligations for minors. Operators must make the AI nature of the service persistently visible or disclose it at the start of each session and again every three hours. Operators may not reward minors for increased engagement, and they must make account and privacy controls available to minors and, where the user is under 13, to parents or guardians.
Two approaches, two loci of accountability
The two laws differ in what they regulate and where accountability sits. Maine regulates the person or system that offers therapy, and it uses licensing law as its enforcement backbone. A chatbot deployed in Maine to provide therapy without a licensed clinician behind it exposes the practitioner or the provider to the existing penalty structure for unlicensed practice.
Nebraska regulates the chatbot itself, and it regulates three distinct behaviours: misrepresenting the service's nature, misrepresenting its credentials, and failing to provide a crisis protocol. The accountability attaches to the operator of the service. The obligations are architectural. They tell the operator how the service must be configured, not only what the service must not do.
The wider landscape
The Future of Privacy Forum is tracking scores of chatbot-specific bills across multiple states and several federal proposals. Missouri's SB 1444 is advancing through an omnibus health care measure. A federal bill introduced by Representative Kevin Mullin in March 2026 would prohibit AI chatbots from impersonating doctors, lawyers, and other licensed professionals. Similar measures have been introduced in states including New York, Pennsylvania, Louisiana, Vermont, and California.
What the April activity makes legible is not simply a growing number of bills but the beginnings of two distinguishable approaches. One treats the chatbot as a provider, and extends professional licensing law to cover it. The other treats the chatbot as a designed system, and imposes positive duties on the operator that configures it.
What the direction of travel reveals
Is it the therapist or the telephone that the law should be regulating when the telephone has started talking back? The question is being answered in practice, on a week by week basis, by state legislatures operating with limited coordination. The answers are converging, which is the more important observation.
The convergence is around the judgment that a conversational AI system which presents itself as a source of professional care or judgment carries risks the existing regulatory architecture was not designed to contain, and that the architecture needs to be extended rather than waived. What varies is the mechanism. Maine uses licensing. Nebraska uses operator duties. The federal proposals so far lean towards impersonation prohibitions. Over a short period, the locus of accountability has begun to move from the person providing a service to the system configuring a conversation.
Opinion: The direction is from licence to architecture
What the April activity makes clear is that the question of whether AI should be permitted to provide mental health services is being overtaken by a more structural question, namely what obligations attach to the operator of a conversational AI system when that system will foreseeably encounter users in psychological difficulty.
Maine's answer is that therapy requires a licensed practitioner, and extending the rule to AI makes the principle uncontroversial. Nebraska's answer goes further. It does not only exclude AI from a regulated domain; it imposes a positive architecture on conversational AI services irrespective of domain: tell the user what you are, do not claim to be a clinician, and have a protocol for the moments when a user reaches for help.
The significance of the Nebraska move is that it treats configuration as a matter of public interest. A conversational AI service is, in this framing, a thing with a design, and the design is accountable. Constitutional rights, including the right to know what one is speaking to and the right not to be deceived about the nature of one's treatment, are protected not by declarations of principle but by obligations attached to the design of the system.
The direction of travel is from licence to architecture. The question that follows is whether the other domains where conversational AI is already operating, education, customer care, workplace support, will be expected to arrive at architecture of their own, or whether they will be allowed to remain at the level of principle until harm has done the persuading.
Sources
Nebraska Legislature, LB 525 (109th Legislative Session)
Capitol Pulse, LB 525: Adopt the Agricultural Data Privacy Act and the Conversational Artificial Intelligence Safety Act
Transparency Coalition, TCAI Bill Guide: Nebraska's AI Chatbot Safety Bills, LB 939 and LB 1185
Maine Legislature, LD 2082: An Act to Regulate the Use of Artificial Intelligence in Providing Certain Mental Health Services
Future of Privacy Forum, 2026 Chatbot Legislation Tracker
Future of Privacy Forum, The Chatbot Moment: Mapping the Emerging 2026 U.S. Chatbot Legislative Landscape
Troutman Pepper Locke, Proposed State AI Law Update, 13 April 2026
Congressman Kevin Mullin, Bill to Stop AI Chatbots from Impersonating Licensed Professionals, March 2026
The contents of this article are for informational purposes only and do not constitute professional, legal, or financial advice.
