FCC Rules on AI-Generated Voice in Robocalls

The FCC’s February 2024 ruling on AI voices is often reported as having made something illegal. It did not. It said that a thing already illegal for thirty years is still illegal when the voice is synthesised, which is a smaller claim and a more useful one, because it tells you exactly which body of consent law you are already inside.

What the FCC ruled, and when

On 8 February 2024 the Commission released a Declaratory Ruling in CG Docket No. 02-278, adopted at the start of that month, holding that calls made with voices generated by artificial intelligence fall within the Telephone Consumer Protection Act’s existing restriction on calls using an “artificial or prerecorded voice”. A declaratory ruling under the Commission’s rules resolves a controversy or removes uncertainty about existing law; it took effect on release, with no compliance date, because on its own account nothing was new.

The immediate context was a robocall sent to New Hampshire voters in January 2024 using a synthesised voice imitating the sitting President and discouraging primary voting. The Commission subsequently pursued forfeiture proceedings against the individual who arranged the calls and reached a settlement with the originating carrier over caller-ID authentication failures. Those are separate proceedings with their own postures, and the outcomes of parallel state criminal charges do not change the regulatory reading.

The consent the TCPA already required

The relevant prohibitions are at 47 U.S.C. § 227(b)(1), with the Commission’s implementing rules at 47 C.F.R. § 64.1200. Two provisions carry most of the weight.

• § 227(b)(1)(A)(iii) prohibits calls using an automatic telephone dialing system or an artificial or prerecorded voice to a mobile number, without the prior express consent of the called party, absent an emergency purpose.
• § 227(b)(1)(B) prohibits calls to a residential line using an artificial or prerecorded voice to deliver a message without prior express consent, subject to exemptions the Commission has adopted for certain non-commercial and non-telemarketing calls.

The distinction to hold on to is between prior express consent and prior express written consent. Under § 64.1200(a)(2) and the definition at § 64.1200(f), a call that includes or introduces an advertisement or constitutes telemarketing needs the written form: a signed agreement, which may be electronic, that clearly authorises calls using an artificial or prerecorded voice to a specified number, discloses that consent is not a condition of purchase. Informational calls need only prior express consent, which can be established more readily.

Because the AI ruling classifies synthesised voice as artificial voice, a campaign that was previously outside the artificial-voice rules because a live agent spoke is inside them the moment the voice is generated — even if the script is dynamic and the conversation responsive. This is the practical trap for AI voice agents: the design intuition is that a responsive agent is more like a live call than a recording, and the rule turns on how the voice was produced, not on how interactive the call is.

The identification rules also apply. Section 64.1200(b) requires an artificial or prerecorded voice message to state clearly at the beginning the identity of the business or individual responsible for the call, and to provide a telephone number during or after the message. A conversational agent that opens by asking a question rather than identifying itself does not satisfy this.

What the ruling does not cover

Being clear about the boundary is more useful than the holding. The declaratory ruling did not create a disclosure duty, did not prohibit voice cloning as such, and did not address text.

• No duty to say the voice is AI. The TCPA requires identification of the caller, not disclosure of how the audio was produced. In August 2024 the Commission issued a Notice of Proposed Rulemaking proposing to define AI-generated calls and to require disclosure of their use; a notice of proposed rulemaking is a proposal, and proposals do not bind anyone.
• No prohibition on cloning a voice. Rights in a person’s voice come from state right-of-publicity law and, in some states, from newer digital-replica statutes — see the California digital replicas page. The FCC regulates the call, not the model.
• Consent is the defence, and it survives. A customer who gave prior express written consent to receive prerecorded marketing calls has consented for TCPA purposes whether the audio is recorded or generated. The ruling did not invalidate existing consents.
• Text messages sit under the same statute but a different analysis. The artificial-voice prong does not apply to SMS; the autodialer prong does, and its scope was narrowed by Facebook, Inc. v. Duguid, 592 U.S. 395 (2021).

Who enforces it, and for how much

The TCPA is unusual among communications statutes in having a private right of action. Section 227(b)(3) allows a person to sue for actual monetary loss or $500 per violation, whichever is greater, and permits a court to treble that to $1,500 where the violation was wilful or knowing. Per violation means per call. Class treatment turns those numbers into the reason TCPA litigation is an industry.

The Commission separately enforces through forfeiture proceedings under 47 U.S.C. § 503, and state attorneys general may bring actions under § 227(g). Voice service providers carry their own obligations under the STIR/SHAKEN caller-ID authentication rules and the robocall mitigation database requirements, which is how the Commission reaches the carrier that originated traffic it did not generate.

What McLaughlin unsettled

In June 2025 the Supreme Court decided McLaughlin Chiropractic Associates, Inc. v. McKesson Corp., holding that in a civil enforcement proceeding the Hobbs Act does not bind a district court to an FCC interpretation of the TCPA. The court must instead determine the statute’s meaning under ordinary principles of statutory interpretation, affording the agency’s view appropriate respect.

That is directly relevant here, because the February 2024 ruling is an interpretation of the phrase “artificial or prerecorded voice”. After McLaughlin, a defendant in a private TCPA suit can argue to a district court that a synthesised voice is not within the statutory phrase, and the court is not obliged to defer to the Commission’s contrary view.

It is worth being precise about how weak that argument looks and how little that matters to the analysis. “Artificial” is not a term of art narrowed to a particular synthesis technique, and the natural reading covers a machine-generated voice comfortably; the Commission’s reading is likely to be adopted by most courts on its own merits. But “likely to be adopted” is a different legal position from “binding”, and it means the answer can now differ between circuits until appellate courts converge. Anyone telling you the question is closed is describing the position before June 2025.

A second reminder that agency rules are not permanent: in January 2025 the Eleventh Circuit vacated the Commission’s “one-to-one consent” rule, which would have required consumer consent to be given to one identified seller at a time, on the ground that it exceeded the agency’s authority to define consent under the statute. The rule had been adopted in December 2023 and never took effect. Between that decision and McLaughlin, the practical guidance for a voice-AI calling programme is to build compliance on the statutory text and documented per-number consent records, which survive changes in agency interpretation, rather than on the current shape of a Commission order.