AI Voice Agent Calls and REPEP/No Llame Coverage

DNC LATAM · Compliance guides

Yes. An AI voice agent placing outbound calls into Mexico or Argentina is covered by REPEP and No Llame the same way a live agent or a predictive dialer is — both registries regulate the contact reaching a registered number, not the technology that generates the agent’s side of the conversation. Swapping a human script for a generative-AI voice model doesn’t open a gap in either registry’s scope.

Why teams assume otherwise

AI voice agents are new enough, and different enough from a traditional predictive dialer, that it’s tempting to treat them as an uncovered category rather than a new instance of an old one. The reasoning usually goes: an AI voice agent isn’t “automated” in the IVR or robocall sense — it holds a real, responsive, often indistinguishable-from-human conversation — so maybe it doesn’t fall under the “automated calls” language that shows up in Argentina’s statute, or under the generic “calls” language in Mexico’s. That reasoning treats conversational sophistication as a compliance-relevant distinction. Neither registry draws that line. We covered the same misreading for robocalls and predictive dialers — the pattern repeats here with a more capable technology attached.

What actually determines coverage

REPEP and No Llame scope a call by what it’s for and which number it reaches, not by how the calling party decided what to say next. An AI voice agent that opens with a sales pitch, qualifies a lead, or tries to schedule a callback is making a marketing contact to a phone number — exactly the kind of contact both registries were built to screen. Whether the words on the call were generated by a language model in real time, read from a fixed IVR script, or spoken by a live agent off a call guide doesn’t change what the number’s owner experiences: an unsolicited commercial call. Argentina’s Registro Nacional No Llame, created by Law 26.951, explicitly names automated calls alongside live ones — see the Argentina Do Not Call list guide for the registry’s full scope, which an AI-driven conversation sits squarely inside. Mexico’s REPEP, run by PROFECO, doesn’t use that exact “automated” framing, but its coverage of marketing calls and messages is broad enough that a conversational AI call to a REPEP-registered number is still a call to a REPEP-registered number.

What’s genuinely different about AI voice agents

The compliance substance doesn’t change, but two operational details do, and they’re worth naming honestly rather than waving away.

First, AI voice agents can run at a volume and speed that outpaces a periodic list scrub even more than a predictive dialer does. A platform that generates and places calls programmatically, often integrated directly via API into a dialer or CRM, can spin up a campaign against a list the moment it’s uploaded — before anyone remembers to scrub it. The dialer-queue problem we described for predictive dialers gets sharper when the thing deciding to dial is also the thing capable of starting a new campaign without a human in the loop to catch a stale list.

Second, an AI voice agent’s opening still has to disclose what a human agent’s opening would. Our post on call opening disclosures covers what a call into Mexico or Argentina needs to establish early: who is calling, why, and how to opt out. None of that changes because the voice on the line is synthetic. If anything, a called party who realizes mid-call that they’re speaking with an AI system is more likely to ask about opt-out, not less — and the opt-out request from an AI-handled call needs to reach the same internal Do Not Call list and dialer-level suppression as one taken by a live agent. An AI voice agent that can’t recognize and log an opt-out request mid-conversation creates the exact compliance gap that agent training for live opt-outs was built to close, just with a model instead of a person failing to catch it.

Where this bites in practice

The failure mode isn’t a legal loophole; it’s an integration gap. AI voice agent platforms are often adopted alongside, or bolted onto, an existing dialer or CRM stack, sometimes by a team that owns the AI platform contract separately from the team that owns DNC scrubbing. That split ownership is where numbers slip through: the AI platform has its own call list, scrubbed once at onboarding or not scrubbed against the current REPEP or No Llame release at all, while the team running traditional campaigns assumes scrubbing happens everywhere calls originate. An auditor reviewing a complaint doesn’t care which system placed the call or who owns which vendor contract — they care whether that number was checked against a current registry release before contact, per the standard covered in DNC compliance certificates.

What this means operationally

Treat every AI voice agent platform the same way you’d treat a new dialer: as a calling channel that needs to sit behind the same scrub, not beside it.

  • Scrub the AI platform’s call list against the current REPEP or No Llame release before the campaign starts, the same as any other outbound list — a separate vendor contract doesn’t mean a separate compliance standard.
  • Build opt-out detection and logging into the AI agent’s conversation flow, not just into a post-call summary a human reviews later, so a spoken opt-out request reaches your suppression list before the next dial attempt, not after a delay.
  • Keep the screening record for AI-placed calls in the same system you’d show an auditor for any other channel — a dated, per-number record that doesn’t change shape because the agent on the call was synthetic.

This is the same standard covered in our contact center compliance guide, applied to a channel that’s new mainly in how fast it can scale, not in what it’s required to check before dialing. DNC LATAM checks a phone number against the current official REPEP or No Llame release at the point any system — AI voice platform included — is about to place the call, with a dated record regardless of what generated the conversation on the other end.