---
title: Can my AI voice agent make and take phone calls?
description: Consent rules for AI calls under the TCPA, recording and transcription consent, AI disclosure, and voice cloning, with a checklist for voice-agent products.
category: Agents
service: agent-product-review
updated: 2026-11-17
---

## Short answer

Yes, with consent and disclosure built in. An AI voice counts as an "artificial voice" under federal robocall law, so outbound calls to consumers generally need the person's prior consent, and sales calls need written consent. Recording or transcribing a call can require everyone's consent in some states. Telling people they're talking to an AI is the safe default. Inbound calls, where people call you, avoid most robocall rules, but the recording and disclosure rules still apply.

## Outbound calls: the TCPA

The Telephone Consumer Protection Act (TCPA) restricts calls that use an "artificial or prerecorded voice." In February 2024, the FCC ruled that AI-generated voices, including voice clones and conversational agents, are artificial voices under the TCPA. That brings AI calls under the same rules as robocalls:

- **Consent before the call.** Calls to cell phones and residential lines with an artificial voice need the called person's prior express consent. Telemarketing calls, meaning calls that promote a product or service, need prior express written consent: a signed agreement, which can be electronic, that clearly authorizes calls from you.
- **Identify yourself.** An artificial-voice call must state the name of the business responsible at the start, and give a phone number during or after the message.
- **Opt-outs.** Telemarketing calls with an artificial voice must offer an automated way to opt out during the call. People can revoke consent in any reasonable way, including by saying so to your agent, and you must honor it promptly. FCC rules allow up to 10 business days.
- **Do Not Call and calling hours.** Telemarketing calls must respect the National Do Not Call Registry and your own internal do-not-call list, and can only be made between 8 a.m. and 9 p.m. in the called person's time zone.
- **Damages.** Violations carry $500 per call, up to $1,500 if willful, and TCPA class actions are common. An agent that makes thousands of calls can create large exposure quickly.

Two developments affect how these rules play out. In *McLaughlin Chiropractic v. McKesson* (June 2025), the Supreme Court held that courts aren't bound by the FCC's interpretations of the TCPA. The AI-voice ruling rests on the statute's plain words, so courts are likely to agree with it, but other FCC positions may be challenged. And in 2024 the FCC proposed rules that would require AI disclosure at the start of each call and in consent agreements. Those proposals haven't been adopted. Building them in now costs little.

**Who is the caller?** If your agent places calls for your users, such as booking appointments or following up with leads, the question is who "initiates" the call. Under FCC precedent, that depends on how much the platform controls the calls: who chooses the numbers, the content, and the timing. Plan as if your company is the caller. Calls to businesses' landlines fall outside the artificial-voice consent rule, but many small businesses answer on cell phones, which are covered.

## Recording and transcribing calls

A voice agent records or transcribes nearly every call it handles. Federal law and most states allow recording when one party to the call consents, and your agent counts as a party. About a dozen states, including California, Florida, Illinois, Maryland, Massachusetts, Pennsylvania, and Washington, require the consent of everyone on the call. When a call crosses state lines, plan for the stricter rule.

AI vendors add a second issue. Lawsuits under California's wiretap law argue that a voice or transcription vendor that can use call content for its own purposes, such as training its models, is a third party listening in, not just a tool. A federal court let claims like these proceed against Google's contact-center AI in 2025, and against the AI note-taker Otter.ai in 2026. Neither case has been decided on the merits, but both shape the safe design:

- Announce at the start that the call is recorded or transcribed, before the substance of the call begins.
- Use voice, transcription, and model vendors whose terms bar them from using your call content for their own purposes, including training. Check those settings, not only the contract.
- Keep recordings and transcripts only as long as you need them, and say how long in your privacy policy.

## Saying it's an AI

Several state laws require telling people they're dealing with an AI, especially in sales, and the pending FCC proposal would require it on calls. An agent that claims to be human, or dodges the question when asked, creates deception risk under the FTC Act and state law. The safe design is to say it in the first sentence: "Hi, this is an AI assistant calling for Northside Dental." See [Do I have to tell users they're talking to an AI?](/guides/telling-users-they-are-talking-to-ai) for the state laws.

## Cloning a real person's voice

Cloning a real person's voice, including your own team's, needs that person's written permission, scoped to how the clone will be used. Several states protect a person's voice as part of their likeness, and Tennessee's ELVIS Act names voice explicitly. Imitating someone to mislead the people you call is deceptive, and potentially fraud. In 2024, the FCC fined the political consultant behind an AI robocall imitating President Biden's voice $6 million.

## Texts

Text messages count as calls under the TCPA, and an AI-written text gets the same treatment as any other. Whether a marketing text legally needs written consent depends on how it's sent, the Do Not Call rules, and state laws such as Florida's that add their own consent requirements, so the safe practice is prior express written consent for marketing texts. Carrier registration rules for business texting apply too.

## Checklist an agent can run against the codebase

- [ ] Find every place the product places outbound calls or texts: dialer integrations (such as Twilio, Vonage, or Telnyx), voice-agent platforms, and scheduled or bulk call jobs.
- [ ] For each, find where consent is recorded before the call: who consented, when, to what, and for which number. Check that sales calls use written consent.
- [ ] Check the agent's opening lines for the business name, a callback number, and an AI disclosure.
- [ ] Check that "stop calling me" and similar requests, by voice or text, update the do-not-call list everywhere, within 10 business days at most.
- [ ] Check that telemarketing calls are screened against the National Do Not Call Registry and the internal list, and limited to 8 a.m. to 9 p.m. in the recipient's time zone.
- [ ] Check that calls are announced as recorded or transcribed before the conversation starts.
- [ ] Check voice, transcription, and model vendor settings for training and retention on call audio and transcripts.
- [ ] Check retention and deletion for recordings and transcripts against the privacy policy.
- [ ] If the product clones voices, find where the voice owner's permission is stored and what uses it covers.

## Sources

- [FCC Declaratory Ruling on AI-generated voices under the TCPA (FCC 24-17, February 2024)](https://docs.fcc.gov/public/attachments/FCC-24-17A1.pdf)
- [FCC telemarketing and artificial-voice rules, 47 CFR 64.1200](https://www.ecfr.gov/current/title-47/chapter-I/subchapter-B/part-64/subpart-L/section-64.1200)

## If this applies to you

An **agent product review** covers a voice agent end to end: consent capture, call scripts and disclosures, recording consent, vendor terms, and the terms that allocate responsibility between you and your users. For one change, such as adding outbound calling to an existing agent, a **pre-build feature check** covers that change before it ships.
