Which rules apply to an AI agent?
Two separate TCPA triggers matter, and AI voice agents usually hit the first:
- Artificial or prerecorded voice (47 U.S.C. 227(b)): AI voices are covered under FCC 24-17. This applies no matter how the number was dialed. An AI agent dialing one CRM contact at a time still needs consent because of the voice.
- Autodialer (ATDS): after Facebook v. Duguid (2021), only equipment that generates numbers randomly or sequentially qualifies. Most list-based dialers don't. State laws in Maryland and Oklahoma reach broader "selection or dialing" systems.
Checklist before an AI agent dials
- Consent on file for every numberMarketing calls to cell phones or residential lines: prior express written consent (signed or e-signed, naming your company and the number, disclosing that consent isn't a condition of purchase). Informational calls to cells: prior express consent.
- Identify the business at the startArtificial-voice messages must state the business's identity at the beginning and give a callback number (64.1200(b)).
- Automated opt-out on marketing callsOffer a key-press or voice opt-out. Honor revocations made by any reasonable means within 10 business days. Don't let the AI keep pitching after "no"; that allegation features in a 2026 class action.
- Scrub against the Do Not Call RegistryAt least every 31 days, and keep an internal DNC list for 5 years.
- Respect local calling hours8am–9pm federally. Florida, Oklahoma, Maryland and Washington are 8am–8pm, Connecticut 9am–8pm, Texas noon–9pm on Sundays. See calling hours by state.
- Recording and transcriptionGet all-party consent in states like California, Florida, Illinois, Pennsylvania and Washington. Under California's CIPA, an AI vendor that can use call data for its own purposes may be treated as an eavesdropper (Ambriz v. Google, 2025).
- State registration and textsTexas requires seller registration and now treats texts as solicitations. Check registration and bond rules where you call.
- Watch AI-disclosure rulesThere's no federal "I'm an AI" requirement yet (FCC 24-84 is still a proposal), but some states, like Utah, require disclosure in certain cases.
The safer design: human handoff
Most AI SDR teams don't have written consent for their outbound list. The pattern that works: the agent does research, selection, timing and the brief, then hands the actual call to a person who dials one number at a time. That avoids the artificial-voice trigger entirely, and you keep the AI's leverage.
// AI SDR: reply says "call me"
await ringahuman.create_call_task({
phone, state, consent: "inquiry",
goal: "Confirm the use case and book a 20-minute demo",
brief: summarize(thread),
webhook_url: "https://agent.example.com/hooks/call-outcome"
})
FAQ
Does the TCPA apply to AI voice agents?
Yes. The FCC's 2024 ruling places AI-generated voices under the TCPA's artificial or prerecorded voice rules, including conversational agents and voice clones.
Do AI SDRs need consent to call leads?
If the AI speaks, yes: prior express written consent for marketing calls to cell phones and residential lines. If a human makes the call from an AI-written brief, TCPA consent generally isn't required, though DNC rules and calling hours still apply.
Can an AI agent leave voicemails?
Only with consent. Ringless and prerecorded voicemail drops are calls under the TCPA (FCC 22-85), and AI-generated voicemails are artificial-voice messages.
Is the platform liable or just the caller?
Usually the caller, but a December 2025 complaint (Lowrey v. Twilio) argues platforms that know about and can prevent violations are liable too. Treat compliance as a product requirement.
Sources
- FCC 24-17 (AI voices) · FCC 24-84 (proposed AI disclosure)
- 47 CFR 64.1200
- Facebook v. Duguid (2021)
- Ambriz v. Google: AI voice and CIPA (Goodwin)
- Lowrey v. Twilio (National Law Review)
General information, not legal advice. Run your call through the TCPA call checker.
Five MCP tools. Free sandbox, no key.