Do You Have to Tell Callers They Are Talking to an AI Agent?

Sometimes legally, always practically. No US federal law currently requires disclosure on inbound calls answered by AI voice agents. But Utah requires it in regulated professions, the EU requires it from August 2026, and if you ever ask a caller "are you a real person?" and your agent lies, that's deceptive practice regardless of state.

Is there a federal law requiring AI disclosure on calls?

Not yet for inbound calls. The FCC has proposed one and has not finalized it. What already exists is a ruling that AI-generated voices count as "artificial" under the Telephone Consumer Protection Act, which matters enormously for outbound calling and barely at all for answering your own phone.

The sequence is worth getting straight, because vendors muddle it. In February 2024 the FCC ruled that AI technologies generating human voices fall inside the TCPA's restriction on "artificial or prerecorded voice," meaning calls using them need the called party's prior express consent. Then in September 2024 the Commission opened a rulemaking proposing to define AI-generated calls and require callers to disclose to consumers when they receive one. That proposal is still a proposal. Nothing in it binds you today.

So the honest federal answer for a business answering its own inbound phone: no disclosure mandate, and a live proposal that could create one. Plan for it rather than waiting for it.

Does California's bot law apply to phone calls?

Almost certainly not — and this is the single most misreported point in every article on this subject. California's B.O.T. Act covers bots operating online, and the statute defines "online" in a way that reads as websites and apps, not telephone calls.

Read the operative language. Business and Professions Code § 17941 makes it unlawful "to use a bot to communicate or interact with another person in California online, with the intent to mislead the other person about its artificial identity" for commercial or electoral purposes. The chapter defines "online" as appearing on any public-facing internet website, web application, or digital application, and defines a "bot" as an automated online account. A phone call is neither.

Two further limits people miss even when discussing the law correctly. First, it only bites where there's intent to mislead about artificial identity — accidental non-disclosure isn't the target. Second, disclosure is a complete safe harbour: "A person using a bot shall not be liable under this section if the person discloses that it is a bot."

I'd still disclose in California, for two reasons that have nothing to do with the B.O.T. Act: California is an all-party consent state for call recording, so you're making an announcement anyway, and the FTC's deception authority doesn't care what channel you used. But if someone tells you the B.O.T. Act requires your phone agent to announce itself, they haven't read the definitions section.

When does Utah require you to say it's an AI?

Utah is the state with a real, operative disclosure duty on commercial AI conversations — but as amended in 2025 it's narrower than it was. For ordinary consumer interactions, you must disclose when a consumer asks. For regulated occupations and high-risk interactions, you must disclose up front, verbally, before the exchange begins.

Utah's AI Policy Act took effect in May 2024 as the first US law of its kind, then SB 226 narrowed it in 2025 so that general-interaction disclosure is triggered by a consumer's request rather than owed proactively, with SB 332 extending the law's life to 2027. The provisions now sit in Utah Code Title 13, Chapter 77. Two practical consequences:

Your agent must answer the question honestly: If a caller asks "am I talking to a person?", the agent has to say no, clearly. An agent that deflects or plays along is the specific failure Utah legislated against. Test this exact phrase before you go live.
If you're in a licensed profession, disclose at the top of the call: The proactive duty attaches to state-licensed occupations and to high-risk interactions involving health, financial, or biometric data.

What changes for healthcare, legal, and financial callers?

The threshold drops sharply, and in California healthcare there's a specific statute — though it has an exemption that covers exactly what most voice agents do. If your agent gives clinical, legal, or financial advice, disclose proactively. If it books appointments, the rules are looser than the panic suggests.

California's AB 3030 is the clearest example. It requires health facilities, clinics, and physician's offices using generative AI to communicate clinical information to include a disclaimer that the message was AI-generated, plus instructions for reaching a human. For audio, the disclaimer must be delivered verbally at the start and the end of the communication.

Now the part that actually determines whether it applies to you. AB 3030 covers "patient clinical information," and the statute expressly excludes "administrative matters, including, but not limited to, appointment scheduling, billing, or other clerical or business matters." It also exempts anything read and reviewed by a licensed provider. A dental or medical voice agent that books appointments and takes messages sits outside the requirement as written.

That's a narrow legal escape and a bad operational plan. A scheduling agent will eventually be asked a clinical question, and the line between "when can you see me" and "should I come in for this pain" is one sentence in a real conversation. Disclose at the top of every healthcare call and you never have to litigate which side of the line a given call fell on.

Does inbound versus outbound change the answer?

Direction of the call matters more than any other factor. Inbound — someone calls you — is lightly regulated. Outbound — you call them with an AI voice — lands squarely inside the TCPA, needs prior express consent, and carries per-call statutory damages. Same software, different legal universe.

On disclosure, there's no federal mandate for inbound today, though an FCC proposal is pending. Outbound is a different matter: the TCPA applies and an AI voice is treated as "artificial." Consent follows the same split — nothing required before an inbound call, but prior express consent before you dial out.

Recording announcements are the one place the two converge: required in all-party consent states either way. The risk profile doesn't converge at all. Inbound exposure is largely reputational, while outbound carries statutory damages per call. So disclose on inbound because it works better, and don't launch outbound without counsel.

The trap is that your vendor sells one product doing both. A voice agent handling missed calls is a low-risk deployment. Flip on outbound recall or reactivation campaigns with the same AI voice and you've changed your regulatory exposure without changing tools. Advantage Labs' voice agents handle inbound answering and booking; if you extend any deployment to outbound dialing, that's the moment to involve a lawyer.

People conflate these constantly. AI disclosure and recording disclosure are different duties from different bodies of law, and the recording one is older, clearer, and more likely to bite you. Roughly a dozen states require all-party consent to record — California, Florida, Illinois, Maryland, Massachusetts, Montana, New Hampshire, Pennsylvania, and Washington are consistently on the list, with a few others varying by source and by how each state's courts have read its statute.

Because most AI voice platforms record and transcribe by default, and because you can't control which state an inbound caller is dialing from, the workable rule is to announce recording on every call to the strictest standard. Then combine it with the AI disclosure in one sentence, which is what the script below does.

What about the EU and callers outside the US?

If you take calls from the EU, disclosure becomes mandatory from 2 August 2026 with no commercial-purpose test and no state-by-state variation. Article 50 of the EU AI Act is the strictest rule in force anywhere and the simplest to comply with.

The obligation: providers must ensure that AI systems intended to interact directly with natural persons are designed so those persons are informed they are interacting with an AI system, delivered "at the latest at the time of the first interaction," clearly and in a manner meeting accessibility requirements. There's an exception where the AI nature is obvious to a reasonably well-informed person — but the Commission's guidance reads that narrowly, and "my agent sounds robotic enough that people can tell" is not a compliance strategy.

Note what's absent: no requirement that you intended to deceive, no requirement that it's a sales call. Direct interaction with a person triggers it. If any part of your call volume comes from Europe, write your script to Article 50 and you've simultaneously satisfied every US rule discussed here.

What should the disclosure actually say?

One sentence, at the top, naming the business, naming the AI, and covering recording. Anything longer gets talked over; anything vaguer fails the "clear and conspicuous" standard that every one of these laws uses.

"Thanks for calling [Business] — you're speaking with our AI assistant, and this call is recorded. How can I help?"

That sentence does four things: identifies the business, discloses the AI plainly, covers all-party recording consent, and hands the call to the caller in under six seconds. Four rules for whatever wording you land on:

Say "AI" or "automated," not a euphemism: "Virtual assistant" and "digital concierge" are the phrasings that fail a clear-and-conspicuous test, because a reasonable caller can hear them as a job title.
Put it in the first utterance: Every regime that specifies timing says first interaction or start of the call.
Answer the direct question honestly, always: Configure the agent so "are you a real person?" returns an unambiguous no. This is the one behaviour Utah's law targets and the one every caller remembers.
Give an exit to a human: AB 3030 requires it for covered communications; everywhere else it's what stops a disclosed AI from feeling like a wall.

Disclosing also just performs better. Callers who know they're talking to a machine ask shorter, more direct questions and get booked faster; callers who work it out at second thirty feel handled and hang up. The client-side write-up on emergency call detection and escalation covers the human-handoff side, which is the other half of making a disclosed agent work.

Frequently asked questions

Do I have to disclose if my AI agent only books appointments?

Under current US federal law, no, and California's AB 3030 explicitly excludes appointment scheduling from its healthcare disclosure duty. Utah still requires an honest answer if a caller asks directly. Most operators disclose anyway, because a scheduling call reliably drifts into questions the agent shouldn't answer undisclosed.

What happens if a caller asks whether they're talking to a robot?

Answer honestly, in every jurisdiction, without exception. Utah's law is built around exactly this trigger, and an agent that deflects is the clearest possible evidence of intent to mislead under California's B.O.T. Act and the FTC's deception authority. Configure it, then test it with the literal phrase.

Is "virtual assistant" enough of a disclosure?

Probably not. Every statute here uses a clear-and-conspicuous or reasonably-informed standard, and "virtual assistant" is a term callers routinely hear as a human role. Use "AI assistant" or "automated assistant." The wording costs you nothing and removes the argument entirely.

Does the disclosure hurt booking rates?

Not in the way operators fear. A small number of callers hang up on a disclosed AI — usually older callers and anyone calling about a problem. Undisclosed agents lose callers too, later in the call and more angrily. Measure your own hang-up rate in the first fifteen seconds before assuming.

Do these rules apply to my after-hours answering service?

Same rules, same analysis — the time of day is irrelevant. What matters is whether the call is inbound, which state the caller is in, whether you're in a licensed profession, and whether the call is recorded. An after-hours AI agent is simply an inbound agent with lower volume and less supervision.

Which US states currently require AI disclosure?

Utah has the operative general commercial duty; California has a narrow online bot law plus a healthcare-specific statute; several other states have passed or proposed rules aimed mainly at companion chatbots rather than customer service. This is the fastest-moving area of AI law — verify your states rather than trusting any published list, including this one.

Different things. Disclosure means telling the caller; consent means getting permission first. Inbound calls generally need disclosure only. Outbound AI-voiced calls need prior express consent under the TCPA, which is a documented opt-in obtained before you dial — not a line read at the start of the call.

Conclusion

Disclosure is the rare compliance question where the strictest standard is also the easiest one to meet: one sentence at the top of the call naming your business, naming the AI, and covering recording. Write it to the EU's Article 50 bar and you've satisfied Utah, California, and whatever the FCC finalizes, without tracking which state each caller dialed from. Then test the literal phrase "are you a real person?" and make sure the answer is an unambiguous no, because that exchange is where both the legal exposure and the caller's trust actually live. Advantage Labs builds inbound voice agents with disclosure, recording language, and human handoff configured from the start. Schedule a consultation with Advantage Labs to review your greeting script before your agent takes its first call.