TCPA consent for AI voice calls after the FCC's 2024 ruling: what's settled and what isn't

The Telephone Consumer Protection Act (TCPA) is the 1991 federal law, at 47 U.S.C. 227, that restricts calls made with an autodialer or an artificial or prerecorded voice. In February 2024 the FCC confirmed that it covers AI voice agents, and three parts of the picture have moved since: how consumers revoke consent, the one-to-one consent rule, and how much weight courts give the FCC. This is where each stands on September 23, 2026, and what a team needs on file to show consent call by call. The FCC votes on September 30, 2026 on a draft order about revocation, and we'll update this post after the vote. It's for compliance leads at collectors, lenders, insurers and healthcare providers, and the engineers who build their agents.

Is an AI voice call an artificial or prerecorded call?

Yes. In a declaratory ruling adopted February 2, 2024 and effective on release on February 8 (FCC 24-17), the FCC confirmed that the TCPA's restrictions on "artificial or prerecorded voice" calls cover "current AI technologies that resemble human voices and/or generate call content using a prerecorded voice." It made no new rule; it applied the existing statute and the rules at 47 CFR 64.1200 to AI calls.

Two lines matter most for voice agents. The TCPA "does not allow for any carve out of technologies that purport to provide the equivalent of a live agent," so a conversational agent gets no exemption for sounding human. And the FCC extended its 2020 soundboard ruling, which held that a live agent choosing prerecorded clips doesn't take a call outside the TCPA, to AI. The artificial-voice prong also stands apart from the autodialer prong, so the narrow autodialer definition in 64.1200(f)(2) doesn't help here.

The ruling's weight changed in 2025. In McLaughlin Chiropractic Associates v. McKesson Corp. (June 20, 2025), the Supreme Court held that district courts in TCPA enforcement suits aren't bound by the FCC's reading of the statute and must interpret it themselves, "affording appropriate respect to the agency's interpretation." So FCC 24-17 persuades a court rather than binding it. The statute's words point the same way, though, and the FCC cited a Ninth Circuit decision describing an artificial voice as "a sound resembling a human voice that is originated by artificial intelligence." Build as if the ruling holds.

What consent does an AI voice call need?

Prior express consent for informational calls to cell phones, prior express written consent for telemarketing, and none for up to three informational calls in 30 days to a residential line, with an opt-out. The purpose of the call and the kind of line decide which:

The cell-phone column comes from 227(b)(1)(A)(iii) and 64.1200(a)(1) and (a)(2); the residential column from 227(b)(1)(B) and 64.1200(a)(3). Health care calls get more depth in HIPAA and AI voice agents in patient access.

Prior express consent has no form requirement. For debt collection, the FCC's 2008 ACA International ruling held that a consumer who gives a creditor a cell number during the transaction that created the debt has consented to calls about that debt at that number. The same ruling puts the burden of proving consent on the creditor and treats a third-party collector's calls as the creditor's own. After McLaughlin, a court reads that ruling for itself too.

Prior express written consent is defined in 64.1200(f)(9): a written agreement, signed (an e-signature valid under federal or state law counts), that clearly authorizes the seller to deliver telemarketing calls using an autodialer or an artificial or prerecorded voice to a stated number. It must tell the signer that the calls are authorized and that signing isn't a condition of any purchase.

The rule AI agents break without anyone noticing is the trigger for written consent: a call that "includes or introduces an advertisement or constitutes telemarketing" (64.1200(a)(2)). A collection agent that improvises an offer for a new card, or a claims agent that pitches an add-on policy, can turn an informational call into a telemarketing call that needed written consent. That's a rule a transcript can check.

Inbound calls sit outside all of this. The consent rules attach to calls a business makes or initiates, and the FCC's ruling speaks of AI technology "that initiates any outbound telephone call." An inbound caller needn't have consented; your agent's callback does.

What must an AI voice message say?

The business's registered name at the start, a callback number, and for some calls an automated opt-out. Every artificial or prerecorded voice message has to meet 64.1200(b), and FCC 24-17 repeats that these requirements reach AI calls:

1. At the beginning, state clearly the identity of the business responsible for the call, using the name under which it's registered to do business with the state.
2. During or after the message, state a telephone number for that business (not the dialer's). For telemarketing and for exempt calls to residential lines, that number must accept do-not-call requests during regular business hours.
3. For telemarketing, and for exempt calls to residential lines, offer an automated, interactive voice or key-press opt-out, with brief instructions, within two seconds of the identification. Using it must add the number to the caller's do-not-call list and end the call at once. A message left on voicemail must give a toll-free number that reaches the same mechanism.

For an agent that improvises, the two-second window is the hard part: a pleasantry and a question before the opt-out already miss it. Collectors face one more collision: Regulation F's voicemail safe harbor wants a business name that doesn't reveal debt collection, which can conflict with the registered-name rule (see our FDCPA and Regulation F checklist for AI collection calls).

No federal rule yet requires an agent to say it's an AI. The FCC proposed AI-call disclosure rules in August 2024 (FCC 24-84), and as of September 23, 2026 nothing from that proposal is in 64.1200. Check state law.

How can a consumer revoke consent, and when do the 2024 rules apply?

By any reasonable means that clearly expresses a wish to stop, under the FCC's February 2024 revocation rules (FCC 24-24), codified at 64.1200(a)(10) and (11) and in effect since April 11, 2025. Using a caller's automated opt-out on a call, replying "stop," "quit," "end," "revoke," "opt out," "cancel" or "unsubscribe" to a text, or using a website or number the caller designated is reasonable per se, and other methods, such as a voicemail or email, create a rebuttable presumption of revocation. Callers must honor a revocation within ten business days and, under the current rule, may not designate an exclusive means of revoking. A consumer who tells your agent "stop calling me" has revoked; the agent's job is to log it where the dialer will see it.

One piece isn't in force. The requirement that a revocation in response to one type of message stops all of a caller's robocalls and robotexts on unrelated matters was waived until April 11, 2026, then extended to January 31, 2027 by DA 26-12. On September 9, 2026 the FCC released a draft order for its September 30 meeting that would let callers apply a revocation only to the category of informational robocalls it was aimed at, and let them designate an exclusive revocation method from the three listed in the rule, provided it's disclosed on the call. A draft isn't a rule and can change before the vote. If it passes as written, an AI agent's in-call opt-out could become the one door consumers are told to use, so it must be announced and must work.

What happened to the one-to-one consent rule?

The Eleventh Circuit vacated it before it took effect, and the FCC took it out of its rules in August 2025. In December 2023 the FCC had changed the definition of prior express written consent so it would cover only one identified seller, for calls "logically and topically associated" with where consent was given, due to take effect January 27, 2025. On January 24, 2025, the Eleventh Circuit vacated that change in Insurance Marketing Coalition v. FCC, holding that it conflicted with the ordinary meaning of "prior express consent." The mandate issued April 30, 2025, and the FCC restored the earlier definition effective August 29, 2025.

What that settles: the one-seller and topic limits are gone from the federal rule. What it doesn't touch: the written-consent requirement for telemarketing robocalls, which comes from a 2012 FCC order the court said was not at issue.

What are the damages for an AI voice call without consent?

Under 227(b)(3), a consumer can sue for actual losses or $500 per violation, whichever is greater, and a court may treble the award, to $1,500, if the violation was willful or knowing. The statute sets no cap, so exposure scales with call volume. Repeat telemarketing calls to numbers on the national do-not-call registry carry a separate right of action under 227(c)(5), and state attorneys general can sue on residents' behalf.

What's settled and what isn't?

The AI-voice ruling, any-reasonable-means revocation and the end of one-to-one consent are settled; revoke-all, exclusive opt-out methods and an AI disclosure rule are not.

How do you prove consent call by call?

With records for each call, not each campaign, because the FCC puts the burden of showing consent on the caller's side. For each outbound AI call, you should be able to produce:

1. The number dialed and whether it was wireless or residential when the call was made.
2. The consent that covers that number: where and when it was given, by whom, to whom, and how it reached the caller.
3. For telemarketing, the signed written agreement with the disclosure text the consumer saw.
4. Every revocation on record for that number, from any channel, with the date received and the date it was honored.
5. A reassigned-numbers check. The safe harbor in 64.1200(m) protects a caller who queried the reassigned numbers database with the consent date and got an erroneous "no."
6. For exempt calls to residential lines, the count of calls to that line in the prior 30 days.
7. From the call itself: the opening identification, the callback number, the opt-out where required, whether the agent introduced an advertisement, and whether a spoken revocation was logged.

Only the last item lives in the transcript and the agent's tool calls. Items 1 to 6 come from consent records, dial logs and database queries, so a transcript-only review can't tell you whether a call was lawful to place. Loops checks that consent was on file before each AI-voice call when it has your dial log and consent records (rule 6.2 in the example collections procedure these guides share); without them it grades what happens inside each call and marks 6.2 not checked, never passed. Our guide to auditing AI voice agent calls against your own procedures covers the in-call half. This is general information, not legal advice.

Next, take a sample of last month's outbound AI calls and assemble items 1 to 7 for each. Where a call can't be backed, fix the record-keeping before the next release, and watch the FCC's September 30 vote, which may change how your agent's opt-out has to work. For collection calls, see how Loops audits collection calls, or start the free 30-day audit for a graded view of your own.