AI debt collection calls in the US: Regulation F and the TCPA, for offshore and Indian teams

Can you use an AI agent for US debt collection calls?

Yes, but the rules were written to stop abusive collection, and they apply to the collector whatever technology places the call. The Fair Debt Collection Practices Act (FDCPA) states its purpose as eliminating "abusive debt collection practices by debt collectors" (15 U.S.C. 1692). Regulation F, the CFPB's rule under that Act, took effect on 30 November 2021 and sets out in detail how a debt collector may communicate (12 CFR part 1006).

Regulation F does not mention AI. Its limits on frequency, timing, disclosures and stop requests therefore apply to an AI agent exactly as they do to a person, and the collector answers for both.

The TCPA is a separate law with a different target: calls made with an autodialler or an artificial or prerecorded voice (47 U.S.C. 227). The Federal Communications Commission (FCC) settled in February 2024 that an AI-generated voice is an artificial voice. Some states add stricter limits on top, so a single AI agent can be answering to three layers of rules on the same call.

Why offshore and Indian vendors are covered too

Teams outside the US sometimes assume these rules stop at the border. In practice they do not. The FDCPA defines a debt collector as anyone who uses interstate commerce, such as phone lines or the mail, in a business whose main purpose is collecting debts. It also covers anyone who regularly collects debts owed to another (15 U.S.C. 1692a(6)).

A business process outsourcing (BPO) firm in Pune dialling US consumers for a US lender is collecting debts owed to another. On our reading, where the vendor sits is not part of the test, so the same rules follow the calls whether they start in the US or in India.

A creditor's own staff collecting in the creditor's name are generally outside the FDCPA. But a creditor that collects under a different name, suggesting that a third party is collecting, becomes a debt collector (1692a(6)). An AI agent that introduces itself under a separate brand can create exactly this problem.

Liability also runs both ways. The FCC has ruled that a seller can be vicariously liable, under federal agency principles, for TCPA violations by third parties calling on its behalf (Declaratory Ruling FCC 13-54, 2013). That ruling was about sellers and telemarketing, but on our reading the same agency logic means outsourcing collections calls does not move the risk away from the US client.

FDCPA statutory damages are up to $1,000 per individual case. In a class action they can reach the lesser of $500,000 or 1% of the collector's net worth (15 U.S.C. 1692k(a)(2)). Because both sides are exposed, contracts should give the client audit rights over dialler logs, consent records and scripts. India's own collection rules are a different regime, covered in RBI's recovery agent rules, and Indian lenders using vendors have outsourcing duties of their own.

How often and when you can call: the 7-in-7 rule and the 8 am to 9 pm window

The two limits that shape an AI dialler most are frequency and timing. Both are built as presumptions rather than flat bans, which matters for how a dialler should be set.

What an AI agent must say on the call and in a voicemail

Once the dialler has cleared a call, the agent's first job is to confirm it is speaking to the consumer. Telling a relative or colleague about the debt is a third-party disclosure problem (12 CFR 1006.6(d)).

In the first communication, the collector must say it is a debt collector attempting to collect a debt and that any information obtained will be used for that purpose (1006.18(e)(1)). Every later communication must say it is from a debt collector (1006.18(e)(2)). This is often called the 'mini-Miranda', after the warning US police give on arrest.

A first-call opening that covers it might run: "This is [Collector], a debt collector. This is an attempt to collect a debt, and any information obtained will be used for that purpose." Calls must also meaningfully disclose the caller's identity (1006.14(g)). Versions of the opening for other markets are in the recording-notice and AI disclosure scripts.

The agent should also give, or trigger, the validation information within five days of the first communication (1006.34(a)(1)). That is the notice telling the consumer about the debt and their right to dispute it.

Regulation F has no rule that the caller must say it is an AI. Under the TCPA rules, every artificial or prerecorded voice message must state the identity of the business responsible at the start and give its telephone number (47 CFR 64.1200(b)). Saying the call uses an automated assistant is the safer practice, and the FCC has proposed making that a rule.

Voicemail is where these duties most easily collide, because a message that says too much becomes a full communication. The note below sets out what a safe message may contain, and the flowchart shows where the disclosures sit in a complete call.

The TCPA and AI voices: consent, landlines and revocation

Regulation F governs what a collector says and when; the TCPA decides whether an AI voice may be used on that line at all. Its answer depends on the kind of line and on consent.

Stricter state limits: Massachusetts and Washington

The federal rules are a floor, not a ceiling: they do not override stricter state rules. The table compares Regulation F with two states whose limits are tighter. Other states have their own licensing and conduct rules, so every state a collector calls into needs mapping before go-live; the calling rules tool is one starting point.

Massachusetts splits its rules by who is calling. The Attorney General's rules cover creditors and their agents and employees, while licensed debt collectors follow Division of Banks rules (209 CMR 18.00). Those were amended with effect from 26 September 2025, and we could not confirm the current text of their call limit, so licensed collectors should check it directly.

Pending FCC changes on AI disclosure and consent

Two parts of this picture are still moving at the FCC. In August 2024 the FCC proposed rules that would define AI-generated calls and require callers to disclose that they use AI-generated voices (FCC 24-84, CG Docket 23-362). No final rule had been adopted as of 24 October 2026. The FCC's regulatory agenda of 14 August 2026 lists the proposal under long-term actions, so it is pending rather than imminent.

The consent-revocation rules are changing too. On 30 September 2026 the FCC adopted an order that lets callers name one or more exclusive ways to opt out, if they disclose them clearly on the call or text (FCC 26-67). The permitted methods are an automated key-press or voice opt-out, a reply using standard opt-out words, or a designated website or phone number.

Under the same order, a revocation made in response to an informational call can be treated as covering only that category of calls. One made in response to telemarketing covers all of the caller's telemarketing robocalls. The order does not say whether collection calls form a single informational category. The ten-business-day deadline to honour a revocation stays for now, although the FCC has asked for comment on changing it.

These changes start 30 days after the order is published in the Federal Register, which had not happened as of 24 October 2026. The FCC says it will announce the exact date in a Public Notice. They replace a wider 'revoke-all' rule that never took effect; the FCC had delayed it to 31 January 2027 (DA 26-12).

Separately, on 24 January 2025 the Eleventh Circuit vacated the FCC's 'one-to-one' consent rule, which would have required marketing consent to name a single seller (Insurance Marketing Coalition v. FCC). That rule concerned telemarketing, so it matters little for pure debt collection calls. It does matter if an AI agent mixes collections with offers such as refinancing. Taking card payments on the call brings in a further standard, explained in PCI DSS rules for card payments on AI calls.

Key dates