ConsentRecording
Research brief

AI Calling Compliance: TCPA, Recording and Disclosure Research

The short version
  • Consent is mainly an outbound question. The FCC confirmed in February 2024 that AI generated voices fall under the TCPA, so a call the business initiates needs prior express consent. The statute sets $500 per violation, and up to three times that if willful or knowing 1,2.
  • The revocation rules are moving. The FCC's 'revoke all' rule was delayed twice and, per a law firm alert, replaced by an order released on 1 October 2026 that was not yet in force when we checked 3,4.
  • Litigation is rising. TCPA class actions went from 539 to 1,052 between the first half of 2024 and the first half of 2025, according to a defense firm's blog quoting WebRecon 5.
  • Recording is a state question, for inbound calls too. One guide counts 11 all party consent states and another counts 13 6,7, and a 2025 ruling let a wiretap claim against an AI call vendor go forward 8.
  • Whether you must say it is an AI depends on where the caller is. The EU duty applies from 2 August 2026, and California and Utah have their own rules 9,10,11,12.
  • This is a research summary and not legal advice. Several items were read through law firm summaries, which the sources list marks.

Selling AI phone agents means selling into a legal map that differs by direction of call, state and country, and some of it changed this month. This page collects the primary texts and the best secondary summaries on US consent rules, call recording laws, AI disclosure duties and litigation volume, and states what each source is and what we could not verify. It is a research summary, not legal advice.

· 14 sources · Published by Convocore, which sells white label AI agents. See method and limits.

$500
per violation is the minimum a person can recover under the TCPA for an unconsented artificial or prerecorded voice call, and a court may raise it up to three times for willful or knowing violations (47 U.S.C. 227(b)(3))1
95.2%
rise in TCPA class actions filed in the first half of 2025 against the first half of 2024: 1,052 against 539, per WebRecon data reported by a defense firm's blog5
11 states
require all parties to consent to recording a phone call, in one consumer guide's count; another guide counts 13 and says some states are situational6,7
2 Aug 2026
is when the EU AI Act duty to tell people they are interacting with an AI system (Article 50) applies, with penalties of up to 15 million euros or 3% of turnover9,10
10 Feb 2025
is when a US federal court refused to dismiss Ambriz v. Google, a class action claiming an AI contact center vendor illegally wiretapped customer calls (pleadings stage, nothing proven)8
31 Jan 2027
was the FCC's delayed compliance date for its 'revoke all' consent rule, which an order released on 1 October 2026 replaces3,4

Start with the direction of the call

Most of the US consent law concerns calls a business initiates. Our reading, not legal advice, is that an AI receptionist answering a customer who calls in is in a different position from an AI agent that places outbound calls, though recording and disclosure rules can still apply to both. The sources below say which rule is about which direction, and where we are reading between the lines we say so.

Which rules touch which kind of call: our reading of the sources, not legal advice2,6,8,9
RuleInbound: a customer calls inOutbound: the business calls out
TCPA prior express consent for an AI generated voice (FCC 24-17)Ruling addresses calls a caller initiatesConsent needed; written consent if telemarketing
Identify the responsible business at the start of the callNot addressed for inbound callsRequired for artificial voice messages
Opt out and consent revocation handlingNot addressed for inbound callsRules are changing, see below
Call recording consent (state wiretap laws)Applies when calls are recorded or transcribedApplies when calls are recorded or transcribed
AI disclosure (EU Article 50, some US states)Depends on where the caller isDepends on where the caller is
Vendor wiretap claims (Ambriz v. Google type)Alleged against the AI vendor in the audio pathAlleged against the AI vendor in the audio path
Source says it appliesUnclear or not addressedNot addressed by the source

Our reading of where each source applies. The FCC ruling is about calls a caller initiates and says nothing on inbound calls. Take legal advice for your own use.

US federal law: the TCPA and the 2024 FCC ruling

The Telephone Consumer Protection Act, at 47 U.S.C. section 227(b)(1), makes it unlawful to make any call, other than one made for emergency purposes or with the prior express consent of the called party, using an automatic telephone dialing system or an artificial or prerecorded voice. Section 227(b)(3) lets a person sue for actual monetary loss or $500 in damages for each violation, whichever is greater. If the court finds the violation willful or knowing it may increase the award to not more than three times that amount 1.

On 8 February 2024 the FCC released a declaratory ruling (FCC 24-17, adopted 2 February 2024) confirming that the TCPA's restrictions on an artificial or prerecorded voice cover current AI technologies that resemble human voices or generate call content using a prerecorded voice. Callers must obtain the prior express consent of the called party before making a call that uses AI generated voice, absent an emergency purpose or exemption, and prior express written consent where the call is advertising or telemarketing. The ruling states that the TCPA does not allow any carve out for technologies that purport to provide the equivalent of a live agent, and that artificial voice messages must state at the start the identity of the business responsible for the call 2.

The ruling concerns calls a caller initiates. It is silent on a customer calling in. State attorneys general and private plaintiffs can enforce the TCPA 1,2.

What the TCPA's damages clause means in dollars1
01
$500
minimum per violation
47 U.S.C. 227(b)(3)
02
Up to $1,500
per violation if willful or knowing
The court may increase the award to up to 3 times
03
$500,000
1,000 calls at the minimum
Illustration, our arithmetic
04
$1.5 million
the same 1,000 calls at the 3x maximum
Illustration, our arithmetic

The first two cards are the statute. The last two are arithmetic, not predictions: real outcomes depend on how courts count violations, defenses and settlements.

How much TCPA litigation is there?

TCPAWorld, a blog run by a class action defense lawyer, reported using WebRecon data that 1,052 TCPA class actions were filed in the first half of 2025 against 539 in the first half of 2024, a rise of 95.2%. It also reported that June 2025 alone had 257 TCPA cases, 202 of them class actions, or 78.6% 5. The blog's publisher sells legal services to defendants, and we did not open WebRecon's own report.

TCPA class actions filed in the first half of each year5
First half of 2024539
First half of 20251,052 (up 95.2%)

Counts as reported by TCPAWorld from WebRecon data, a defense firm's blog. We did not open WebRecon's report.

June 2025: TCPA cases filed, by type5
Class actions202 (78.6% of cases)
Other TCPA cases55
257 total less 202 class actions, our subtraction

AI vendors and wiretap claims: Ambriz v. Google

A newer risk is that the AI vendor, not just the business, is sued over call interception. In Ambriz v. Google, a putative class action in the US District Court for the Northern District of California about Google's Cloud Contact Center AI, the court denied Google's motion to dismiss on 10 February 2025. According to a Goodwin alert of 26 February 2025, the court held the plaintiffs plausibly alleged a wiretap under section 631(a) of the California Invasion of Privacy Act, that Google read the content in transit because it learned of it before or at the same time as the business customer, and it followed the 'capability test' from Javier v. Assurance IQ. Under that test a third party software provider can violate the law if its technology has the capability to use the data for its own benefit, whether or not it actually did 8.

This was a ruling at the pleadings stage, so the allegations are unproven. Goodwin notes the law is uncertain given a split between the capability test and a narrower 'extension' test. We could not confirm what happened to the case after that ruling, and a commentator suggests the standing question may now be raised on appeal. The alert recommends that customers disclose the use of third party AI voice technology and obtain the consent of both parties 8.

How the vendor liability theory works, as the court described it8
01
Call
A customer phones a business that uses an AI contact center
02
Audio
The vendor's software hears the call in transit, before or as the business does
03
Capability
Under the 'capability test' a vendor can be a wiretapper if it can use the data for its own benefit
04
Claim
A class action under CIPA 631(a), allowed to proceed on 10 Feb 2025 and unproven

A single pleadings stage order from a law firm alert. The courts are split between the capability test and a narrower extension test.

Telling people it is an AI

The EU AI Act's Article 50(1) requires providers to design AI systems that interact directly with people so that those people are informed they are interacting with an AI system, unless that is obvious to a reasonably well informed, observant and circumspect person. The article applies from 2 August 2026 9. A Jones Walker article of 16 July 2026 reports that the Digital Omnibus delayed the high risk rules to 2 December 2027 and 2 August 2028 but left most Article 50 transparency obligations on their original date, with a grace period to 2 December 2026 only for marking of generated content in systems already on the market. It reports Article 50 breaches fall in the tier of up to 15 million euros or 3% of worldwide annual turnover, whichever is higher, with the lower figure applying to qualifying SMEs. Those are ceilings, not automatic penalties 10.

In the US there is no single federal duty to disclose an AI voice on an inbound call that we found. California's bot disclosure law, SB 1001, makes it unlawful to communicate with a person in California online with the intent to mislead them about the bot's artificial identity in order to incentivize a purchase or sale or influence a vote, and a clear, conspicuous disclosure avoids liability under it, per law firm summaries 11. Utah's Artificial Intelligence Policy Act requires a person who uses generative AI to interact with a consumer to disclose, clearly and conspicuously, that it is generative AI if asked, with fines of up to $2,500 per violation, and has since been amended to focus on high risk interactions 12. These two were read through law firm summaries only, and the Utah text has changed, so check the current statutes.

Disclosure is also a business question. In SurveyMonkey's December 2025 poll of 2,017 US adults, 89% said companies should always offer the option to speak to a person 13. In a 2019 randomized trial on more than 6,200 customers, disclosing that an outbound sales caller was a chatbot reduced purchases by more than 79.7% 14. Legal duties and commercial effects can pull in opposite directions.

EU AI Act: what applies when, as reported by a law firm in July 20269,10
  1. 2 Aug 2026
    Article 50 applies
    People must be told they are interacting with an AI system unless it is obvious. Most transparency duties kept this date after the Digital Omnibus.
  2. 2 Dec 2026
    Grace period for content marking ends
    Applies only to machine readable marking of generated content in systems already on the market before 2 August 2026.
  3. 2 Dec 2027
    High risk rules, stand-alone systems
    Delayed from 2 August 2026 by the Omnibus.
  4. 2 Aug 2028
    High risk rules, AI in regulated products
    Delayed by the Omnibus.

The law firm says publication of the Omnibus in the Official Journal should be confirmed before relying on the delayed dates. Article 50 breaches carry ceilings of up to 15 million euros or 3% of worldwide turnover.

Disclosure pulls in two directions13,14
US adults who say companies should always offer a human option89%
SurveyMonkey, December 2025, 2,017 US adults
Fall in purchases when a chatbot disclosed itself before an outbound sales call79.7% or more
Marketing Science field experiment, 2019, more than 6,200 customers

Different questions and populations, shown together because the legal duty to disclose and the commercial effect of disclosing can conflict.

AI disclosure rules, as reported2,9,10,11,12
JurisdictionRuleSource
European UnionArticle 50(1): people must be told they are interacting with an AI system unless obvious. Applies 2 Aug 2026AI Act text, Jones Walker 9,10
CaliforniaSB 1001: unlawful to mislead about a bot's identity to incentivize a sale or vote; clear disclosure avoids liabilityLaw firm summaries 11
UtahDisclose generative AI use if asked; amended to focus on high risk interactions; up to $2,500 per violationLaw firm summaries 12
US federal (outbound)Prior express consent for AI generated voice calls; identify the responsible business on the callFCC 24-17 2

What the evidence does not show

  • This page is a research summary and not legal advice. Laws differ by state and country, change frequently and depend on facts. Talk to a lawyer before running outbound AI calls or recording calls.
  • The 1 October 2026 FCC order was read through one law firm alert and not the order, and it was not in force when we checked. Earlier dates come from law firm summaries, not FCC documents 3,4.
  • The litigation counts come from a defense firm's blog quoting WebRecon, and we did not open WebRecon's report 5.
  • State recording laws were read through vendor guides that disagree on the count, not through the statutes. We could not open the California Penal Code pages, which blocked our access 6,7.
  • The Ambriz ruling is a single pleadings stage order read through a law firm alert. We could not confirm the case's later history, and the law on whether vendors can be wiretappers is unsettled 8.
  • The California and Utah AI disclosure rules were read through law firm summaries, and Utah's statute has been amended. Other states have their own rules that we have not covered 11,12.
  • We cover the US and the EU only. Rules in the UK, Canada, Australia and elsewhere differ, and we have not researched them.
  • We found no study that measures how often AI calling or AI call recording leads to enforcement, apart from litigation counts for the TCPA generally.

Applying this as an agency

The sources suggest a short checklist to take to a lawyer for each client, not a legal conclusion. Is the call inbound or outbound? For outbound, where does consent come from and how is it recorded and revoked 1,2,4? Are calls recorded or transcribed, and are any callers in all party consent states 6? Does the agent say it is an AI, and what do the laws of the caller's country require 9,10? Which vendors sit in the audio path, and have they been told 8?

On Convocore each client has its own sub-account with its own agent, greeting and call logs, so disclosure wording and retention can be set per client. Software cannot decide the legal questions above. Get advice for each client's states and countries, and keep records of consent.

White label AI platform →Pricing →

Method and limits

  • Every number on this page comes from a source we opened and read on 9 October 2026, except where a source note below says we could only read a search summary.
  • Sources are labelled primary (read at the original publisher), commissioned (paid for by a party with a commercial interest in the result), secondary (a report of the original) or forecast. Law firm alerts are labelled secondary, and statutes and agency rulings primary.
  • Where sources disagreed or a figure could not be traced to an original, we left it out. Dates are given wherever the source gave them.
  • Convocore sells white label AI phone agents and has a commercial interest in this market. We have included risks that argue for caution. If you find an error, write to support@convocore.ai and we will correct it.

Sources

  1. Legal Information Institute, Cornell Law School. Current US Code, read 9 October 2026.
    We read sections 227(b)(1) and 227(b)(3). The page is long and we read the first 100,000 characters of 132,090.
  2. US Federal Communications Commission. Adopted 2 February 2024, released 8 February 2024.
    We read the full ruling and the Chairwoman's statement. It concerns calls a caller initiates and does not discuss inbound calls.
  3. Burr & Forman LLP. January 2026.
    A law firm summary. It states the 6 January 2026 order and the new date but does not say which other revocation provisions are in effect. The 7 April 2025 first delay and 11 April 2026 date come from other law firm summaries in search results.
  4. Manatt, Phelps and Phillips LLP. 5 October 2026.
    A law firm client alert on an order released 1 October 2026. We did not find or read the order. It was not yet published in the Federal Register when the alert was written.
  5. TCPAWorld (Troutman Amin, LLP), citing WebRecon. 25 July 2025.
    A defense lawyer's blog. We did not open WebRecon's own report. We quote only the half year class action counts, the 95.2% rise and the June 2025 split as the blog states them.
  6. Rev. 1 November 2025.
    A transcription vendor's consumer guide, not a statute. It lists 11 all party states, 3 situational states and 38 one party states plus DC, and says to follow the strictest law for cross state calls and consult an attorney.
  7. Koncert support. Undated.
    A software support page. We read only a search summary, which reports 13 states including Connecticut and Michigan.
  8. Goodwin Procter LLP. 26 February 2025.
    A law firm alert on Ambriz v. Google (N.D. Cal., order of 10 February 2025). We did not read the order. The alert states no damages figures. We could not confirm the case's later history.
  9. EU Artificial Intelligence Act, text at artificialintelligenceact.eu (Future of Life Institute). Regulation (EU) 2024/1689; applies 2 August 2026.
    An unofficial but widely used copy of the Regulation's text. We read Article 50(1), (2) and (4). Check EUR-Lex for the official text.
  10. Jones Walker LLP. 16 July 2026.
    A law firm article. It says publication of the Omnibus in the Official Journal should be confirmed before relying on the delayed high risk dates. We did not read the Omnibus text.
  11. Perkins Coie LLP. 2018 to 2019.
    We read only search summaries of this and similar law firm pieces, not the statute (California Business and Professions Code section 17940 and following).
  12. Davis Wright Tremaine LLP. April 2024, with later amendments.
    We read only a search summary. The summary also reports that the law was later amended to focus on high risk interactions, so the 2024 description may be out of date.
  13. SurveyMonkey. Fieldwork 10 to 11 December 2025.
    2,017 US adults from a non-probability online panel, weighted to Census data, margin of error 2.5 points. SurveyMonkey sells survey software.
  14. Marketing Science, vol. 38, no. 6. Published online 20 September 2019.
    We read the abstract and introduction. Randomized field experiment, more than 6,200 customers, highly structured outbound sales calls, 2019 technology.

Questions

Does the TCPA apply to AI voice agents?

In a declaratory ruling released 8 February 2024 (FCC 24-17), the FCC confirmed that the TCPA's restrictions on an artificial or prerecorded voice cover current AI technologies that resemble human voices or generate call content using a prerecorded voice. Callers need the called party's prior express consent. The ruling concerns calls a caller initiates. This is not legal advice.

What are the TCPA penalties?

47 U.S.C. 227(b)(3) lets a person recover actual monetary loss or $500 in damages for each violation, whichever is greater, and a court may increase the award to not more than three times that amount if the violation was willful or knowing.

Is the FCC revoke all consent rule in force?

Not as originally written. The FCC delayed it to 31 January 2027, and a law firm alert reports that an order released 1 October 2026 replaces it, taking effect 30 days after Federal Register publication. Check the order itself.

Which states require all parties to consent to call recording?

Rev's November 2025 guide lists 11: California, Delaware, Florida, Illinois, Maryland, Massachusetts, Montana, Nevada, New Hampshire, Pennsylvania and Washington, and treats Connecticut, Michigan and New Jersey as situational. Another guide counts 13. Check the statutes.

Do I have to tell callers they are talking to an AI?

It depends on where. EU AI Act Article 50(1) applies from 2 August 2026 and requires people to be told they are interacting with an AI system unless obvious. California's SB 1001 and Utah's AI Policy Act have their own disclosure rules, as summarized by law firms. Take legal advice for your states and countries.

Related

AI voice agent research brief →AI receptionist research brief →White label AI for MSPs →White label AI answering service →White label overview →