Is AI Cold Calling Legal: TCPA Rules for AI Voice Agents
AI voice agents count as artificial voices under the TCPA, so outbound AI calls need consent, often in writing. A practical guide to US, UK and Kenya rules, penalties and the November 2026 changes.

AI cold calling is legal in the United States, the UK and Kenya, but only under conditions that most off-the-shelf voice agent setups do not meet out of the box. US law treats an AI-generated voice as an artificial voice, which puts every AI outbound call under the Telephone Consumer Protection Act and its consent rules. This guide maps what the rules actually require, what changed in 2025 and 2026, and where AI cold calling is simply the wrong tool.
We have written before about building AI cold calling assistants. This article is about the legal side: what you must have in place before one of those agents dials a number.
An AI voice is an artificial voice under US law
The Federal Communications Commission settled the threshold question in February 2024. In a declaratory ruling adopted on February 2, 2024, the FCC confirmed that AI technologies capable of generating human voices fall under the TCPA's restrictions on artificial or prerecorded voice calls. The ruling pointed to court decisions describing an artificial voice as one "resembling a human voice that is originated by artificial intelligence."
The practical consequence: every rule written for robocalls applies to your AI voice agent. It does not matter that the voice sounds natural, that the conversation is interactive, or that a language model generates the words in real time. If the voice on the call is generated, the robocall regime applies.
That single fact drives everything else in this article.
United States: consent is the whole game
The consent tiers
The TCPA makes it unlawful to make any call "using an automatic telephone dialing system or an artificial or prerecorded voice" to a cell phone without the prior express consent of the called party, with narrow exceptions (47 U.S.C. § 227(b)(1)(A)).
FCC regulations then split consent into tiers (47 CFR § 64.1200):
- Informational AI calls such as appointment reminders or delivery notices: prior express consent. A customer who gives you their number in the course of business has usually given it.
- Telemarketing AI calls, which includes sales outreach: prior express written consent. That means a signed agreement, with a clear and conspicuous disclosure, authorizing marketing calls using an artificial or prerecorded voice to a specific number.
Notice what is missing from the second bullet: anything resembling cold outreach. If you bought a list, scraped numbers, or pulled contacts from a data broker, you do not have prior express written consent. Pure cold calling to US consumers with an AI voice is effectively banned. The lawful version of AI outbound calling is calling people who opted in, or warm follow-up with existing customers.
The courts are cutting back the FCC's consent rules
Two recent decisions matter for planning.
In January 2025, the Eleventh Circuit vacated the FCC's one-to-one consent rule in Insurance Marketing Coalition v. FCC. That 2023 rule would have required consumers to consent to calls from one identified seller at a time, limited to topics logically associated with the interaction that prompted the consent. It was aimed at lead-generation sites that collected one form fill and sold it to dozens of callers. The court held the FCC exceeded its statutory authority, and the rule never took effect.
In February 2026, the Fifth Circuit went further in Bradford v. Sovereign Pest Control of TX. A customer gave his cell number on a pest-control service agreement and later received prerecorded calls about renewal inspections. The court held he had given prior express consent, and that the TCPA requires no more than that for any type of prerecorded call, affirming judgment for the caller even without written marketing consent.
Read these cases carefully before relaxing your standards. The FCC's written-consent regulation is still on the books, and most circuits have not rejected it. Plaintiff lawyers file TCPA suits where the law is friendliest to plaintiffs. The safe operating standard remains: written, specific, documented consent for AI telemarketing calls.
Calling hours, do-not-call lists and identification
The FTC's Telemarketing Sales Rule adds a second layer on top of the TCPA (FTC compliance guide):
- Calling hours: no outbound telemarketing calls before 8 a.m. or after 9 p.m. local time at the called person's location, absent prior consent. Your dialer must resolve time zones, not run on server time.
- Do-not-call lists: scrub against the National Do Not Call Registry and maintain your own entity-specific list.
- Existing customers: the established business relationship exemption covers live calls only, not automated calls or robocalls. An existing relationship does not authorize AI voice calls.
- No abandoned calls: when a person answers, the call must connect properly, and you must transmit caller ID.
What changes on November 25, 2026
A new FCC rule on revocation of consent takes effect on November 25, 2026 (Federal Register, October 26, 2026). Three operational changes:
- Callers may interpret a revocation request as applying only to the specific category of informational robocalls it was directed at. A customer can stop payment reminders without killing fraud alerts.
- Callers may designate an exclusive means to revoke consent.
- Financial institutions get a broader exemption to alert customers about account fraud.
The direction is friendlier for legitimate callers, but the baseline is unchanged: when someone revokes consent, you stop. Your AI calling stack needs revocation handling that works mid-conversation, because people will tell the agent "stop calling me" out loud. If the agent misses that, every later call is a fresh violation.
State law: California adds an AI disclosure duty
California's AB 2905 amended Section 2874 of the Public Utilities Code. Existing law already required anyone using an automatic dialing-announcing device to state the nature of the call, identify the business represented, and ask whether the person consents to hear the message. AB 2905 adds one more requirement: the announcement must tell the person if the message uses an artificial voice.
One detail worth noting: under California law, violating a Public Utilities Commission order or requirement is a crime. AI disclosure in California is not a best practice. It is a statutory obligation with teeth, and other states are moving the same way. If you call nationally, building the disclosure into your opening script is the simplest way to stay ahead of the patchwork.
United Kingdom: stricter than the US
The UK rules live in the Privacy and Electronic Communications Regulations, enforced by the Information Commissioner's Office (ICO telephone marketing guidance).
The key provision is Regulation 19: you must not make an automated marketing call unless the person has specifically consented to receive this type of call from you. The ICO is explicit that general consent for marketing, or even consent for live calls, is not enough. The consent must specifically cover automated calls. Every automated call must state your name and give a contact address or freephone number, and you must transmit your number.
Live calls are looser: you can call numbers not registered with the Telephone Preference Service, provided you screen against the TPS and CTPS registers and keep your own do-not-call list.
For an AI voice agent, the conservative reading is that it counts as an automated call. That makes cold AI calling to UK consumers effectively closed unless they opted in to automated calls specifically.
Kenya: data protection law reaches cold outreach
Kenya has no robocall statute, but the Data Protection Act 2019 gives the Office of the Data Protection Commissioner jurisdiction over marketing uses of personal data, and the ODPC uses it.
In December 2025, the ODPC ordered Platinum Credit Limited to pay a complainant KES 400,000 for unsolicited promotional SMS messages and calls, even though he had never shared his data with the lender (ODPC Complaint No. 1951 of 2024). The company argued the caller was an outside agent. The regulator established that the caller represented Platinum Credit, which increased its exposure. The ODPC also recommended prosecuting the directors for giving false or misleading information during the investigation, an offence carrying up to KES 3 million in fines, ten years in prison, or both.
Two lessons for anyone running AI outbound calling from or into Kenya. First, how the agent sounds matters less than where the number came from: consent to use the data is the foundation. Second, lying to the regulator is worse than the original violation. Consent duties also follow the call itself: the call recording consent rules in Kenya for AI voice agents cover what you must disclose before the agent starts recording.
What the penalties look like in practice
Run the numbers before a plaintiff's lawyer does it for you:
- TCPA private actions: $500 per violating call, or actual monetary loss if greater, and courts can treble that to $1,500 per call for willful or knowing violations. A 5,000-call AI campaign without consent carries $2.5 million in statutory damages before class certification is even argued.
- FCC fines: the TRACED Act added a civil forfeiture of up to $10,000 per call for intentional violations, on top of the FCC's base penalties and separate from anything private plaintiffs recover.
- Kenya: compensation orders per complainant, plus prosecution exposure for obstructing the ODPC.
Class actions multiply the per-call figure by everyone on the list. This is why the plaintiff bar watches AI calling product launches.
A pre-launch checklist for AI outbound calling
- Classify every campaign. Informational or telemarketing? The answer sets the consent tier.
- Capture the right consent. For telemarketing AI calls in the US: written, signed, specific, with the required disclosure. Store the proof with timestamps.
- Scrub your lists. US: National DNC plus your own list. UK: TPS and CTPS. Kenya: your suppression list of objectors.
- Disclose in the first ten seconds. AI voice, business name, purpose of the call, and how to opt out.
- Build revocation handling into the conversation. "Stop calling me" must trigger an immediate opt-out across every campaign. That means the agent has to hear and act on interruptions mid-sentence, which is a harder engineering problem than it sounds; how interruption handling really works in voice agents walks through the mechanics.
- Respect calling hours by the callee's time zone, not yours.
- Transmit caller ID and make sure callbacks reach a working channel.
- Tune answering machine detection so you never abandon a live answer. The accuracy and cost trade-offs are covered in our guide to answering machine detection for voice AI.
- Log everything: consent records, scripts, recordings, opt-outs.
- Have counsel review before launch and after every rule change. The November 25, 2026 revocation rule is a good trigger for a review.
Common mistakes that create liability
- Treating a purchased lead list as consent. Even before the vacated one-to-one rule, a lead sold to many buyers rarely equals prior express written consent for robocalls from your company.
- Assuming existing customers are fair game for AI calls. The FTC is explicit: the established business relationship exemption covers live calls, not automated calls or robocalls.
- Letting the AI improvise past the approved script. Your consent covers specific calls. A model that freelances into new topics can wander outside it.
- Missing mid-call opt-outs. With the revocation rule taking effect on November 25, 2026, this becomes the highest-risk gap in most setups.
- Calling outside the 8 a.m. to 9 p.m. window because the dialer runs on server time.
When AI cold calling is the wrong choice
Skip the AI caller entirely when:
- You have no consent trail. No configuration fixes that. Pick another channel first.
- You sell high-trust, high-ticket services to enterprises. A synthetic voice on a cold call signals the opposite of what you want. A human rep with AI research support converts better.
- You operate in a regulated sector without a compliance function. Banks, insurers and healthcare providers need documented governance around outbound AI before the first call, which is its own project and part of our enterprise AI work.
- Your volume is low. Under a few hundred calls a month, humans are cheaper and safer.
Frequently asked questions
Is AI cold calling legal in the US? Yes, conditionally. The FCC treats AI-generated voices as artificial voices under the TCPA, so telemarketing calls need prior express written consent and informational calls need prior express consent. Pure cold calling to consumers with an AI voice is effectively banned.
Do I have to tell people the caller is an AI? In California, yes. AB 2905 requires the announcement to disclose when a message uses an artificial voice. Federal law does not yet impose a blanket AI disclosure rule, but disclosure is cheap insurance and good practice everywhere.
Can I use an AI voice agent for cold calls in the UK? Only with consent that specifically covers automated calls. General marketing consent is not enough under PECR Regulation 19, and the ICO enforces it.
What happens when someone asks the AI to stop calling? You must honor it. Consumers can revoke consent, and rules taking effect November 25, 2026 let callers define an exclusive revocation channel. Build "stop calling me" detection into the agent itself.
Is AI cold calling legal in Kenya? The Data Protection Act 2019 requires consent for marketing use of personal data, and the ODPC has ordered compensation for unsolicited marketing calls and texts. Consent and honest records are the baseline.
How much can violations cost? Private TCPA plaintiffs can recover $500 per call, trebled to $1,500 per call if the violation was willful or knowing. The FCC can separately impose forfeitures of up to $10,000 per call for intentional violations. One non-compliant campaign can create seven-figure exposure.
Where to go from here
The pattern across all four jurisdictions is the same: consent first, disclosure up front, opt-outs honored instantly. That is an engineering problem as much as a legal one, and it is solvable. We build outbound voice agents as AI Employees with consent capture, do-not-call screening, disclosure lines and revocation handling built into the call flow. If you want AI outbound calling that survives contact with a regulator, talk to us.
About AI Agents Plus Editorial
AI automation expert and thought leader in business transformation through artificial intelligence.



