No US federal rule makes a business announce that its inbound phone agent is an AI. But three other regimes reach your clients, one started applying on August 2, and the state rule most likely to bite rewards disclosing anyway. What an AI agency actually needs to know.
Lumina9 min read
No federal rule in the United States makes a business announce that the voice answering its phone is an AI. The FCC ruling everyone cites covers outbound robocalls, not a homeowner who dials a plumber at 9pm and reaches an agent. That answer is correct as far as it goes, and it is the wrong thing to plan around. Three other regimes do reach your clients, one of them started applying on August 2, and the state rule most likely to bite rewards doing more than it asks.
Nothing here is legal advice. We sell an agency stack, not legal work, and these rules move fast enough that an attorney should read anything you put in a client contract.
Does US federal law require you to disclose an inbound AI receptionist?
No. The FCC's Declaratory Ruling FCC 24-17, adopted February 2, 2024 and released February 8, confirmed that calls made with AI-generated voices count as "artificial" under the TCPA, which means they need the called party's prior express consent. That is the ruling behind every "AI robocalls are illegal now" headline.
Read the text and the scope is tight. Paragraph 9 closes with this: "These requirements are applicable to any AI technology that initiates any outbound telephone call using an artificial or prerecorded voice to consumers." The whole document is keyed to initiating calls. Inbound is never addressed. A customer dialing a business and reaching an AI is neither permitted nor prohibited anywhere in FCC 24-17.
Carry that split into every sales call. Inbound answering, the thing most of your clients want, sits outside the ruling. Outbound calling, including the missed-call recovery and follow-up sequences that make the ROI math work, sits inside it and needs consent before the agent dials with a synthetic voice. Sell outbound AI calling to a client with no consent record and you have sold them a problem.
One line in paragraph 6 will get quoted at you. The TCPA "does not allow for any carve out of technologies that purport to provide the equivalent of a live agent." In context it aims at outbound callers claiming their bot is a person, but it is the sentence a plaintiff's firm would use to argue broader reach.
Which state rules actually apply to a business answering its own phone?
Utah's is the one to know. It is specific, it is in force, and it rewards doing more than the minimum.
S.B. 226, effective May 7, 2025, says a supplier using generative AI in a consumer transaction "shall disclose to the individual that the individual is interacting with generative artificial intelligence and not a human," but only "if the individual asks or otherwise prompts the supplier about whether artificial intelligence is being used." The statute narrows it again: the question "must be a clear and unambiguous request to determine whether the interaction is with a human or with artificial intelligence."
So a general business gets a reactive duty. Wait to be asked, then answer honestly.
Two things change that. Regulated occupations come first. Utah defines those as occupations regulated by the Department of Commerce that require a license or state certification, and for them disclosure has to be prominent and up front when the use counts as a "high-risk artificial intelligence interaction." That covers collecting health, financial or biometric data, or giving advice someone could reasonably rely on for a significant personal decision. Timing is spelled out: "verbally at the start of a verbal interaction." Dental, med spa, chiropractic, law, accounting, insurance. A good share of the trades we write packs for.
The part people miss is the safe harbor. Under section 13-75-104, a business is not subject to enforcement under the disclosure section at all if its AI "clearly and conspicuously discloses" at the outset and throughout the interaction that it is generative AI, is not human, or is an AI assistant. Fines otherwise run to $2,500 per violation, enforced by the Division of Consumer Protection.
The arithmetic is not close. The reactive standard saves your client one sentence at the top of a call. The proactive standard buys them out of the enforcement question. Tell your clients to disclose.
What changed in Europe on August 2
Article 50 of the EU AI Act started applying on August 2, 2026. It requires that AI systems interacting directly with people be built so those people are informed they are dealing with an AI, unless that is obvious.
The "obvious" exception is narrower than it sounds. The Commission's guidance says it "should be interpreted in a restrictive manner, given that it deprives people of transparency," and the test runs against an average person who is reasonably well-informed, circumspect and observant. A voice agent good enough to sell is, by construction, not obvious. Timing is explicit too. People must be told "from the start of the first interaction in a clear and distinguishable manner." Fines reach 15 million euros or 3% of worldwide turnover.
The duty under Article 50(1) falls on the provider of the system rather than the business deploying it, which for most of you means it falls upstream. That does not make it irrelevant. Provider status applies whether or not the provider sits in the EU, and non-EU providers are covered where their system's output is used in the EU. If you are reselling to anyone with European operations, who counts as the provider once you rebrand and configure a system is the kind of question that gets argued after the fact. Worth a call with counsel before you sign that client.
The healthcare wrinkle nobody reads correctly
California's AB 3030, chaptered as Chapter 848, Statutes of 2024, makes health facilities, clinics and physician offices attach a disclaimer when generative AI produces patient communications about clinical information. For audio, the disclaimer "shall be provided verbally at the start and the end of the interaction." There is an exception when a licensed human provider reads and reviews the communication.
The part that gets misquoted in both directions is the scope. The statute defines "patient clinical information" as information relating to a patient's health status, and expressly excludes administrative matters "including, but not limited to, appointment scheduling, billing, or other clerical or business matters."
So a dental AI receptionist that books cleanings and quotes a price band is outside it. The same agent the moment it answers "does this sound like something I should come in for" is not. That line is where your prompt design becomes a compliance decision, and it is the best argument for the guardrails you should be writing into every medical vertical anyway. More on that trade in how to sell AI systems to dental practices.
What does not apply to you, despite the headlines
California's AI Transparency Act, SB 942 as amended by AB 853, also became operative on August 2, 2026, and drew a pile of coverage. Morgan Lewis has a clear summary.
It almost certainly does not touch you. It regulates covered providers, meaning entities that create a generative AI system publicly accessible in California with more than one million monthly visitors or users, and it governs provenance marking on generated images, video and audio. By its own terms it does not apply to AI-generated text.
I mention it because a prospect who reads two newsletters will bring it up. Knowing which rules are theirs beats having a compliance slide.
Where this gets uncomfortable
The rules above are a snapshot, not a settled position. Utah's underlying policy act carries a 2027 repeal date. California has pending bills that would strip the million-user threshold out of its transparency act entirely. Any fixed compliance claim you write into a client contract will age badly. Write a process instead. We disclose at the start of every interaction, and we review it annually.
The second thing is more awkward for anyone selling this. Disclosure costs conversion, a little. An agent that opens with "you are speaking with an automated assistant" loses a fraction of callers who would have stayed on with a voice they took for a person. Nobody in this business likes saying that out loud. The trade is still worth making, because the other side of it is your client's license and your name on the system that cost it.
A prospect who asks you to build an agent that hides what it is has not made a configuration request. They have told you who they will blame later.
What to actually put in the script
One sentence at the top, in the business's own voice, naming what the caller reached and offering a way to a human. "Hi, this is the automated assistant for Northside Heating and Air. I can book, quote and answer questions, and I can get you to a person any time you ask."
That sentence clears Utah's safe harbor, meets Article 50's start-of-interaction timing, and handles AB 3030's front-end rule for audio. One line, three regimes. Then make the route to a human actually work, because a disclosure with no way out is what regulators care about.
Every demo on our site already says the agent is a demo and that bookings inside it are simulated, and that line stays visible on published demos under your brand. It is the one Lumina-adjacent thing a prospect always sees, and we explain why in what your prospects actually see when you sell white-label AI.
FAQ
Does my client need consent to run outbound AI follow-up calls?
If the call uses an artificial or prerecorded voice, FCC 24-17 says yes, prior express consent, and prior express written consent when the message is telemarketing or carries an advertisement. Texts and inbound answering are a separate analysis. Have counsel review this one rather than taking any blog's word for it, including ours.
Is a chat widget treated the same as a phone agent?
Not always. Utah's timing rule for regulated occupations splits verbal from written, requiring disclosure "verbally at the start of a verbal interaction" and in writing before a written one. AB 3030 sets different placement rules for written, continuous chat, audio and video. Same substance, different placement.
Can I write "complies with AI disclosure law" in my proposal?
Do not. Your client operates the agent, not you, and the rules differ by state, by trade and by licensure. Sell the capability and the default behavior, the same way our FAQ states what a demo discloses. Leave the legal conclusion to their attorney.
Does any of this apply if my client is a plumber or a roofer?
Less than for a dental practice, since the regulated-occupation and healthcare rules turn on licensure and clinical content. The general consumer-transaction duty still applies wherever a state has one, and the trade licenses some contractors hold can pull them into scope. The one-line disclosure covers it either way. That is why we treat it as the default, not the exception.
Next step
Open the trade pack for the vertical you sell into most and rewrite the agent's opening line so it names itself in the first sentence. Then call it, ask for a human, and see whether the handoff works. Twenty minutes, and it clears the only compliance question a prospect is likely to raise on your next call.
Want to hear what a disclosed agent sounds like first? Talk to the live demo on our homepage and ask it to get you a person.
