If no board licenses you, you have probably been reading AI regulation news as somebody else's problem. That was a reasonable read. It stopped being true this week, and it did not stop in a way anybody was watching for.
Signal 1
PSA, ten statutory regulators and the Accredited Registers Collaborative commit to shared AI principles
On September 17 the UK Professional Standards Authority signed a joint statement with ten health and care regulators and the Accredited Registers Collaborative. They agreed to write one shared set of rules for how the people on their registers may use AI. It follows the National Commission into the Regulation of AI in Healthcare, which published its recommendations a week earlier. The Accredited Registers are the part to notice. That program exists for the jobs no law regulates. Counselors, psychotherapists, nutritional therapists, sports therapists and public health practitioners can all work without registering with anyone, so instead of licensing them the Authority approves the voluntary registers they join. Nothing has been written yet. This is a commitment to write it, and the people it will cover were named in the room.
Sources: Nursing and Midwifery Council, September 17, 2026 | Community Care, statement of intent detail | Professional Standards Authority, About Accredited Registers
What this means for you
Look at who signed, not at what was announced. Ten bodies that license people signed it. So did the body that speaks for people nobody licenses. Both groups are getting the same rules at the same time, and the second group got there without a single law being passed. The rule found them through a register they joined voluntarily. If any board or certifying body holds your name on a list, that list is now the most likely way an AI rule reaches you, and it does not need a legislature to get there.
The Pattern
One signal this week, and it is the one worth the whole issue. Rules normally travel through licensing. That is why a practitioner nobody licenses has been able to read AI regulation as somebody else's business for three years and be right. What happened on September 17 is that a rule started traveling down a different road. It went through a register, which is a list somebody joins voluntarily, which no legislature has to approve and no board has to enforce. Registers are not the only road of that kind. A platform's terms, a client contract, an insurer's conditions and a certifying body's code of ethics all reach a practitioner without asking whether anyone licensed them. The exposure was never that these rules are unfair. It is that they arrive somewhere nobody thought to watch, and most people in this position are still watching the legislature.
What To Watch
Two bills sit on the California Governor's desk with a September 30 deadline. Both of them regulate what an unlicensed person may do with AI, which is not a thing any state has tried to regulate directly before.
AB 1979 would deem any business offering a health care chatbot to consumers a provider of health care under the Confidentiality of Medical Information Act, and would separately prohibit using an AI tool to guide an unlicensed individual through a clinical function that requires a license. For a practice running an intake bot, a symptom questionnaire, or an assistant that discusses health with clients, the first half moves you into a confidentiality category most unlicensed practitioners have never been in and are not set up for. The second half is the sharper one, because being walked through clinical reasoning by a capable tool is a large part of how unlicensed practitioners currently use AI at all. Watch it because it is the first serious attempt to regulate the combination of an AI and an unlicensed person as its own category, rather than regulating the tool and the person separately, and because the sponsor is a nurses association. The boundary is being drawn by a licensed profession, around everybody else.
Source: https://leginfo.legislature.ca.gov/faces/billNavClient.xhtml?bill_id=202520260AB1979
SB 903 would bar unlicensed AI therapy services outright, hold licensed professionals to administrative and supplementary AI use only, require consent before AI records or transcribes a session, and give the Department of Consumer Affairs the power to investigate and fine. It passed the Senate unanimously and the Assembly 71 to 4. Those margins matter more than the text does. A bill that close to unopposed is not politically contested, which is how language spreads to other states within a session or two. If any part of your work sits near emotional or behavioral territory, which is to say stress, habits, burnout, or what a person is carrying around a diagnosis, the line between that and therapy is about to be drawn in statute rather than by how you describe yourself. Four states already ban AI therapy. This would add the largest one, and unlike the others it arrives with an agency that can levy penalties. Enforcement is the part that changes behavior.
Source: https://leginfo.legislature.ca.gov/faces/billNavClient.xhtml?bill_id=202520260SB903
Federal: HTI-5 would remove the AI model card requirement from certified health records systems. A model card is the disclosure that states what a tool was trained on, how well it performs, and where it fails. No unlicensed practice runs a certified records system, so this one reaches you through what goes missing rather than through what it requires. If the federal requirement is deleted, the only model card standard still standing is a voluntary registry run by CHAI, which also referees adoption and runs vendor-funded deployments. The next time you ask a vendor how you are supposed to know their tool works, the answer becomes a document produced under rules written by an organization those vendors fund. Watch it because it decides whether accountability for health AI is written by a public body or a private one. Still unpublished as of this week.
Source: https://www.federalregister.gov/documents/2025/12/29/2025-23896/health-data-technology-and-interoperability-astponc-deregulatory-actions-to-unleash-prosperity
One Thing You Can Do This Week
Open the code of ethics or standards document from whoever certified you. Search it for three words: technology, artificial, and disclosure. Then look at the date it was last revised. Five minutes tells you which of two situations you are in. Either an AI expectation has already been written for you and you have not read it, or that document is old enough that one is coming. Both of those are workable. The one that is not workable is not knowing, and that is where most people are right now.