AI Voice Disclosure Laws for Insurance Agents (2026)

An insurance lead picks up, hears a voice that sounds like a person, and starts answering questions about their Medicare Advantage plan. Nobody on that call has said the word “AI.” Is that legal? The honest answer is: it depends entirely on where that lead’s phone number lives, and most agents adopting AI calling right now have never checked. Federal law requires you to identify your business at the start of an artificial-voice call and get consent before you dial, but it does not, on its own, require you to say the words “you’re talking to AI.” A handful of states have gone further and written their own AI-specific disclosure rules. Most states haven’t touched the question at all.

This article works through exactly what the FCC’s TCPA ruling on AI voices actually requires, which states currently have their own AI voice disclosure statute, which widely repeated claims about those laws turn out to be wrong, what a Medicare-specific disclaimer has to do with any of this, what a violation costs, and the full method for building a compliant disclosure into a calling script, whether you do it by hand or with a platform that builds it in.

The Short Answer: It Depends on Where Your Lead Lives

There is no single federal law that says “an AI voice agent must announce itself as AI before speaking to a consumer.” What exists instead is a patchwork: one federal ruling that folds AI voices into an existing consent-and-identification framework, two state statutes as of this writing that add an AI-specific disclosure requirement, and a Medicare-specific disclaimer rule that has nothing to do with AI at all but gets confused with it constantly.

What the FCC’s TCPA Ruling Actually Requires

On February 8, 2024, the FCC released a declaratory ruling, FCC 24-17, in CG Docket No. 23-362, confirming that AI technologies which generate or clone human voices fall within the TCPA’s existing restriction on calls using an “artificial or prerecorded voice.” That’s the entire holding. The ruling doesn’t create a new rule for AI specifically; it applies an old rule, written in 1991, to a new technology.

Here’s what that means in practice, straight from the ruling’s text. Callers using an AI-generated or cloned voice must obtain the called party’s prior express consent before initiating the call, and written consent specifically if the call introduces an advertisement or constitutes telemarketing. Separately, under 47 CFR Section 64.1200(b)(1), which the ruling explicitly applies to these calls, “all artificial or prerecorded voice telephone messages shall, at the beginning of the message, state clearly the identity of the business, individual, or other entity that is responsible for initiating the call.” And if the call is an ad or telemarketing, it must offer a compliant opt-out method under Section 64.1200(b)(3).

Read that identification requirement again: it’s about disclosing who is calling, not what is calling. A script that opens with “Hi, this is Sarah calling on behalf of Riverside Insurance” satisfies the identity-disclosure requirement whether Sarah is a licensed agent or an AI voice model, because the rule was written for prerecorded human voices decades before generative AI existed and the FCC simply confirmed it now also covers AI-generated ones. Nowhere in the ruling’s text, background, or ordering clause does the FCC require the caller to state that the voice is AI-generated. That’s a materially different requirement, and it’s one several vendor blog posts blur together when they summarize this ruling.

The Two States That Currently Require AI-Specific Disclosure

As of this writing, two states have statutes that go beyond the federal identification requirement and specifically address disclosing that a voice or interaction is AI-generated. They work very differently from each other.

Utah: Disclose If Asked, Except for Licensed Professions

Utah Code Section 13-2-12, enacted by Senate Bill 149 and effective May 1, 2024, sets up two separate obligations. The general rule, in subsection (3), applies to “a person who uses, prompts, or otherwise causes generative artificial intelligence to interact with a person in connection with any act administered and enforced by the [Division of Consumer Protection].” That division enforces Utah’s Consumer Sales Practices Act, which covers deceptive acts in consumer transactions broadly, and a sales call offering insurance coverage to a Utah resident is a consumer transaction. Under this subsection, if the person you’re calling asks whether they’re talking to a human, you have to tell them, clearly and conspicuously, that they’re interacting with generative AI, not a human.

The stricter rule sits in subsection (4): “a person who provides the services of a regulated occupation shall prominently disclose when a person is interacting with a generative artificial intelligence in the provision of regulated services,” and that disclosure has to happen proactively, verbally at the start of the exchange. But “regulated occupation” is defined narrowly in the statute as one “regulated by the Department of Commerce that requires a person to obtain a license or state certification.” Utah licenses insurance producers through the Utah Insurance Department, a separate agency from the Department of Commerce. Read literally, that means the strict, proactive, no-one-has-to-ask-first disclosure duty likely doesn’t extend to a standard insurance sales call the way it would to, say, a licensed contractor or a certified appraiser working under Commerce’s umbrella. The general if-asked duty in subsection (3) still applies, because it isn’t limited to regulated occupations at all; it applies to any consumer-facing use of generative AI the Division of Consumer Protection could reach.

California: Disclose Before the Message Plays, for Automated Dialing Devices

California’s Public Utilities Code Section 2874, amended by Assembly Bill 2905 and effective January 1, 2025, governs “automatic dialing-announcing devices,” a term that predates AI and originally covered equipment that dials a number and plays a prerecorded message. AB 2905 added a requirement that before the message plays, an unrecorded, natural-voice announcement must, among other things, “inform the person called if the prerecorded message uses an artificial voice.” The statute defines “artificial voice” as one “generated or significantly altered using artificial intelligence.” A violation is enforced under Public Utilities Code Section 2876, which caps the fine at $500 per violation, or the California Public Utilities Commission can order the device’s phone service disconnected.

There’s a genuine scope question worth naming honestly rather than glossing over: California’s statute is written around a device that plays a prerecorded message, which is squarely the older robocall model. Whether a modern, real-time conversational AI voice agent, one that generates its responses live during the call rather than playing a fixed recording, falls inside “automatic dialing-announcing device” as defined isn’t spelled out with total clarity in the statute’s text. The safest posture for an agency calling California numbers with any AI voice technology is to treat the disclosure requirement as applicable and build it in regardless of how a court might eventually parse the device definition; the fine is small, but a dispute over statutory interpretation isn’t a fight worth picking for the sake of skipping one sentence in a call script.

The State That Doesn’t: Correcting the Texas SB 140 Myth

Several blog posts and vendor compliance guides claim Texas Senate Bill 140 requires AI voice technology to be disclosed within the first 30 seconds of a call. We pulled the enrolled bill text directly from the Texas Legislature’s own site to check. It isn’t there.

SB 140, effective September 1, 2025, does real things: it amends the Texas Business & Commerce Code’s definition of “telephone solicitation” to explicitly cover text, graphic, and image messages, not just voice calls, and it ties a violation of the state’s telephone-solicitation and telemarketing chapters (Chapters 302, 304, and 305) directly to the Texas Deceptive Trade Practices Act, giving consumers the DTPA’s private right of action for a violation. None of that is an AI voice disclosure requirement. It’s a definitional-scope and enforcement statute, full stop, and several secondary summaries of it in circulation describe registration fees, bonds, and calling-hour windows that don’t appear in the enrolled bill text itself.

This is exactly the kind of misattribution the insurance lead-gen and AI calling space is thick with right now: a real bill number, a real effective date, and a specific-sounding requirement that simply isn’t in the text. SB 140, the bill repeatedly cited for a “30-second AI disclosure” rule, doesn’t contain one, and it doesn’t contain the registration and bonding requirements some of the same sources attach to it either.

Medicare Calls Carry a Completely Different Disclosure Requirement

If you work Medicare leads, there’s a disclosure rule that already applies to you regardless of any AI-specific law, and it’s easy to confuse with one because both get called “the disclosure requirement” in casual conversation. CMS’s Medicare Communications and Marketing Guidelines require every Third Party Marketing Organization, a category that covers agents, agencies, and lead generators marketing Medicare Advantage or Part D plans on a carrier’s behalf, to give a standard disclaimer on marketing calls: that the agent doesn’t offer every plan available in the area, and where to go for the full picture. That disclaimer discloses the limits of what you’re selling. It has nothing to do with whether an AI or a human is speaking.

The two obligations stack rather than substitute for each other. A Medicare marketing call placed by an AI voice agent to a Utah or California number needs the TPMO disclaimer because it’s a Medicare marketing call, and it may also need the state’s AI disclosure language because of where the lead lives and what technology placed the call. Treating “we already say the TPMO disclaimer” as covering an AI disclosure obligation, or the reverse, is the kind of gap that shows up during an audit, not before one. We’ve covered CMS’s TPMO rules and the 2027 marketing changes in more depth elsewhere; the compliance disclaimer at the end of this piece links to what’s specific to AI voice disclosure, not the full Medicare marketing framework.

What a Violation Actually Costs

None of these per-violation numbers looks large next to a single sale. They stop looking small at volume. An agency dialing a few hundred Utah or California numbers a week without a disclosure line built in isn’t risking one $500 fine; it’s risking $500 multiplied by however many calls a regulator or plaintiff’s attorney decides to count, on top of whatever separate TCPA or state telemarketing exposure already exists for other reasons. The fix costs nothing: one sentence, said every time.

How to Build a Compliant Disclosure Into Your Script

None of this requires buying anything. Here’s the method, whether you do it by hand with a human team or with a platform that automates it.

1. Open every call by identifying your business, every time, everywhere. This satisfies the federal identification requirement under 47 CFR Section 64.1200(b)(1) regardless of what state you’re calling, and it costs one sentence: your business name and, if it’s a warm follow-up, what the person requested.
2. Add an explicit AI identification line for Utah and California numbers, and default to it everywhere else. “You’re speaking with an AI assistant for [business name]” is short, honest, and satisfies California’s proactive rule and Utah’s stricter reading in one line, without requiring you to maintain two different scripts by state.
3. Program a direct, honest answer for “am I talking to a real person?” This single conditional response satisfies Utah’s general if-asked duty on its own, independent of which subsection technically governs your specific license type.
4. Get prior express consent before you dial, in writing if the call could be read as telemarketing. This is the federal baseline under the FCC’s ruling and predates any state AI law; it doesn’t go away because a state disclosure rule doesn’t apply to you.
5. Keep the TPMO disclaimer and the AI disclosure as two separate lines in a Medicare script, not one merged sentence. They satisfy different requirements and auditors will look for both independently.
6. Log which disclosure language played on which call, tied to the lead’s state. If a regulator or a lead ever asks what was actually said, “the script says X” is weaker than a per-call record showing X was actually delivered.
7. Review this list against the current statute text at least twice a year. State AI law is moving fast; a rule that doesn’t exist today, the way Texas’s doesn’t, could exist next legislative session, and a rule that exists today could change scope.

What the Disclosure Actually Sounds Like

Reading a statute is one thing; writing the actual line a caller says is another. Here’s what a compliant opening looks like at each level, in plain language rather than legal paraphrase.

Federal baseline, every state: “Hi, this is calling from [Business Name] about the quote you requested.” That’s the full identification requirement under 47 CFR Section 64.1200(b)(1): who’s calling and why. Nothing about AI is required here.

Utah, if asked: Caller: “Am I talking to a real person?” Response: “You’re speaking with an AI assistant for [Business Name], not a live agent. I can transfer you to one if you’d like.” Short, direct, and it satisfies the general if-asked duty in Utah Code Section 13-2-12(3) without needing to resolve whether the stricter regulated-occupation rule technically applies to your license type.

California, proactively: “Hi, this is an AI assistant calling on behalf of [Business Name] about the quote you requested.” Stated before any substantive back-and-forth, this satisfies Public Utilities Code Section 2874’s requirement to disclose the artificial voice before the message plays, and it doubles as the federal identification line, so it isn’t an extra step, just a slightly longer first sentence.

Medicare marketing call, any state: Add the TPMO disclaimer as its own sentence, separate from any AI disclosure line: “We don’t offer every plan available in your area. Any information we provide is limited to the plans we do offer. You can also contact Medicare.gov or 1-800-MEDICARE for your full range of options.” Merging this with an AI disclosure sentence makes the call longer and harder to audit against either requirement independently; keep them as two distinct lines even if they run back to back.

None of these lines requires a legal background to say correctly, and none of them costs more than a few seconds of call time. The only real engineering problem is making sure the right one fires automatically based on the lead’s state and the type of call, rather than depending on a live agent remembering which of four scripts applies to which number.

Where TheAffordableAI Fits

We’re not a law firm and this article isn’t legal advice for your specific setup. What we run is the calling layer underneath whatever disclosure policy you land on: an AI voice agent that opens every call with business identification, can be configured with an explicit AI-disclosure line for the states that call for one, and logs every call’s script and disposition to the contact record automatically.

What “Good” Looks Like

Where This Is the Wrong Thing to Worry About

Be straight about the limits here. If you’re not calling Utah or California numbers, there’s currently no state-specific AI voice disclosure statute you need to build around; the federal identification and consent rules apply everywhere, but the “say it’s AI” requirement, as of this writing, doesn’t extend past those two states in any form we could verify. Building an elaborate 50-state AI disclosure matrix for a law that exists in two states is effort spent on the wrong problem. And no disclosure script, however well built, fixes a consent problem upstream of it; if the number was never validly consented to be called in the first place, the fix is in how the lead was sourced, not in what the AI says once it’s already dialing.

State AI legislation is also moving quickly enough that this list will be incomplete within a year. Treat the state count in this article as a snapshot verified against the actual statute text on the date it was written, not a permanent map, and re-check before you assume a state you’re calling today still has no AI voice disclosure rule next year.

Most of what circulates about AI voice disclosure law in insurance calling is either a federal rule stretched to mean something it doesn’t say, or a state requirement repeated from a blog post that never checked the bill text. The actual landscape, verified against primary sources, is narrower and more specific than the rumor mill suggests: one federal identification-and-consent rule that applies everywhere, two states with their own AI-specific disclosure statute, and a Medicare disclaimer rule that’s about something else entirely. Build a script around what the law actually says, review it twice a year, and you’ve covered more ground than most agencies calling today.