By Alexander Flores, Debt Buyer & Triton Financial Solutions, LLC · Level 2: Intermediate · Last updated: October 30, 2026
AI tools are now pitched to every collection agency and debt buyer: voice agents that place calls, chatbots that take payment plans, models that write emails, and scores that decide which accounts to work first. The pitch is usually about cost and liquidation. The question operators ask us is simpler: what are we actually allowed to do?
The short answer is that there is no federal “AI rule” for debt collection. The FDCPA, Regulation F, the TCPA and the FTC Act apply to an AI tool exactly as they apply to a human collector, and in a few places, like AI-generated voices, regulators have said so directly. On top of that, a handful of states have their own bot and automated-decision laws, and one of the most watched, Colorado’s, was blocked in court and then replaced this year.
This explainer walks through each rule, what it means for AI in practice, and the controls we’d want in place before turning a tool loose on a purchased portfolio. Every source below was checked live on October 30, 2026.
Is AI legal in debt collection?
Yes. No federal law bans AI in debt collection, but the FDCPA, Regulation F, the TCPA and the FTC Act apply to AI calls, texts, emails and chats just as they apply to human collectors. Some states add disclosure duties.
That framing matters because it puts the risk in the right place. Regulators do not need a new rule to act against a bot that misrepresents a balance, calls too often or ignores an opt-out. The debt collector deploying the tool owns its conduct. “The vendor’s model did it” is not a defense under any of the statutes below.
Which federal rules apply to AI collection tools?
Four federal sources do most of the work: the FDCPA, the CFPB’s Regulation F, the TCPA as interpreted by the FCC, and the FTC Act’s ban on unfair or deceptive practices. Each applies by channel and conduct, not technology.
| Rule | What it says (plain English) | What it means for AI tools |
|---|---|---|
| FDCPA, 15 U.S.C. 1692d(6) | Placing calls without meaningful disclosure of the caller’s identity is harassment | An AI voice agent must identify who is calling, just like a person |
| FDCPA, 15 U.S.C. 1692e, and Reg F § 1006.18 | No false, deceptive or misleading representations; initial communications must disclose the collector is attempting to collect a debt | Generated scripts, emails and chat replies need the required disclosures and cannot misstate balances, consequences or legal status |
| Reg F § 1006.14(b) | Presumption of compliance if calls about a particular debt stay within 7 in 7 days, and none within 7 days after a phone conversation about that debt | Voice agents and dialers must share one call counter with human collectors |
| Reg F § 1006.6 | Inconvenient times and places, third-party disclosure limits, and a clear opt-out in every email and text | AI-generated emails and texts need consumer-local send times, opt-out language and suppression when a consumer opts out |
| Reg F § 1006.100 | Keep records evidencing compliance until three years after the last collection activity on the debt | Keep transcripts, recordings, prompts or scripts in use, and model outputs sent to consumers |
| TCPA, 47 U.S.C. 227, and FCC Declaratory Ruling FCC 24-17 | Calls using an “artificial or prerecorded voice” need prior express consent absent an emergency or exemption | AI-generated human voices count as artificial voices, so consent rules apply |
| FTC Act, 15 U.S.C. 45 | Unfair or deceptive acts or practices are unlawful | Applies to how AI is used with consumers and to how vendors market AI to agencies |
Sources: statutes and regulations linked above, checked October 30, 2026. This table summarizes; read the full text before configuring any system.
What did the FCC say about AI-generated voices?
In a Declaratory Ruling released February 8, 2024, the FCC confirmed that the TCPA’s “artificial or prerecorded voice” restrictions cover AI technologies that generate human voices. Those calls need the called party’s prior express consent absent an emergency or exemption.
The ruling, FCC 24-17 in CG Docket No. 23-362, is the clearest federal statement on AI in collections calls, even though it isn’t collections-specific. For an agency, three practical points follow:
- An AI voice agent is not a live agent. If a synthetic voice is speaking, treat the call as an artificial-voice call under the TCPA, even if the conversation is interactive and sounds human.
- Consent becomes a data field you must be able to prove. For purchased portfolios, that means knowing what consent the original creditor obtained, for which number, and whether it was revoked before or after the sale. If the data tape doesn’t carry consent history, a voice-AI campaign on that file is a risk you are taking blind.
- Wireless and residential lines have different rules. The TCPA treats calls to residential lines and to wireless numbers differently, so your number-type data has to be accurate before a campaign starts.
Does an AI chatbot or voice agent have to say it is AI?
No federal collection rule requires an AI disclosure, but a bot cannot misrepresent who it is or the debt. California requires bot disclosure in some online commercial interactions, and Colorado’s new laws add disclosure duties from January 1, 2027.
Here is where the state picture stands as of today:
- California. Business and Professions Code § 17941, operative since July 1, 2019, makes it unlawful to use a bot to communicate online with a person in California with intent to mislead them about its artificial identity in order to incentivize a commercial transaction. A clear and conspicuous disclosure that it is a bot avoids liability. Whether a given collection chat falls inside that language is a question for counsel, but disclosure is the simple way to stay out of it.
- Colorado. The 2024 Colorado AI Act (SB 24-205) was set to apply on June 30, 2026. According to the Colorado Attorney General’s AI page, Senate Bill 26-189, signed in May 2026, repeals and reenacts those provisions as an automated decision-making technology law that takes effect January 1, 2027. A Legal 500 summary reports that a federal court order of April 27, 2026 blocked enforcement of the 2024 law, and that the replacement takes a narrower, notice-based approach. The Attorney General’s page also references a separate chatbot law, House Bill 26-1263, and proposed rules filed August 11, 2026.
Our recommendation is operational, not legal: tell consumers when they’re talking to an automated system, and give them a fast way to reach a person. It costs almost nothing, it lines up with where state law is heading, and it removes a whole category of deception arguments under FDCPA § 1692e.
What happened to the CFPB’s AI-related guidance?
On May 12, 2025, the CFPB withdrew 67 guidance documents, including its 2022 and 2023 circulars on adverse action notices and complex algorithms. The underlying statutes and regulations, including the FDCPA and Regulation F, did not change.
The withdrawal notice, 90 FR 20084, lists Consumer Financial Protection Circular 2022-03 (adverse action notices for credit decisions based on complex algorithms) and Circular 2023-03 (adverse action notices and the proper use of the Regulation B sample forms) among the items withdrawn. It also withdrew Circular 2022-04 on insufficient data protection or security, which matters for any AI vendor handling consumer files. The notice says the Bureau will not prioritize enforcement of withdrawn guidance while its review continues, and also that the withdrawal “is not necessarily final.”
Two cautions for anyone reading that as a green light. First, those circulars applied mainly to credit decisions under ECOA and Regulation B, not to most collection activity, and the statutes they interpreted remain law. Second, the same notice points out that state attorneys general can enforce parts of the Consumer Financial Protection Act and that the FTC shares authority over unfair and deceptive practices. Less federal guidance doesn’t remove state enforcement or private FDCPA lawsuits.
The CFPB’s June 6, 2023 report, Chatbots in consumer finance, is still published on its site as research. Its central concern is still a useful design test: deficient chatbots that block access to live, human support can lead to law violations and consumer harm. In collections, that maps directly to disputes, hardship requests and cease-communication requests a bot fails to recognize.
Where does AI create the most compliance risk in collections?
The highest-risk uses are AI voice calls without documented consent, generated messages that misstate debts or consequences, bots that miss disputes or cease requests, and contact engines that bypass Reg F call caps or time-of-day rules.
| AI use case | Main rules in play | Risk level (our view) | Key control |
|---|---|---|---|
| AI voice agent placing outbound calls | TCPA artificial-voice consent; FDCPA § 1692d(6); Reg F § 1006.14 | High | Documented consent by number, shared call counter, identity disclosure |
| Chatbot or AI agent on the payment portal | FDCPA § 1692e; Reg F § 1006.18; California § 17941 | Medium | Bot disclosure, approved responses on balances and fees, human handoff |
| AI-drafted emails and texts | Reg F § 1006.6 opt-out and timing; § 1006.18 disclosures | Medium | Templates locked by compliance; opt-out inserted automatically |
| Inbound call summarization and QA scoring | Recording consent laws vary by state; § 1006.100 retention | Low to medium | Recording notices; retain transcripts with the account |
| Propensity-to-pay and work-queue scoring | FTC Act § 5; state ADMT laws (Colorado from 2027) | Low to medium | Documented model inputs; no protected-class proxies; periodic review |
| Dispute and hardship detection | FDCPA § 1692g; Reg F § 1006.38 | Low if used as a safety net | Route flagged accounts to humans, never auto-close |
Risk levels are DebtMarket’s editorial judgment for planning purposes, not a regulatory classification.
The pattern in that table is consistent. AI is lowest-risk when it helps humans spot problems, like flagging a dispute in a transcript, and highest-risk when it speaks to consumers on its own or makes contact decisions that skip the rules engine.
How should debt buyers and agencies govern AI tools?
Treat every AI tool as a channel on the same compliance stack as your human collectors: one rules engine, documented consent, locked disclosures, human handoff, full records and vendor oversight. Test it before production and audit it after.
Here is the control checklist we’d use before any AI tool touches consumer accounts:
- Inventory every AI use. List each tool, which channel it uses, which accounts it touches and who approved it. Many agencies already run AI inside a dialer, CRM or letter vendor without calling it that.
- Route all contact through one rules engine. AI calls, texts and emails must count against the same Reg F § 1006.14 call caps and § 1006.6 time windows as human contact. Ask vendors to show this in a demo, not describe it.
- Prove consent before voice AI. Store consent source, date and number for every artificial-voice call. For purchased paper, confirm what the seller’s tape and media actually contain.
- Lock what the model is allowed to say. Use approved language for balances, fees, settlement offers, credit reporting and legal status. Free-form generation on those topics is where § 1692e problems start.
- Disclose automation and offer a person. Tell consumers when they’re dealing with an automated system and give them a one-step path to a human, especially for disputes, hardship and cease requests.
- Keep the records. Save transcripts, recordings, the script or prompt version in use, and the messages actually sent, for the full § 1006.100 retention period.
- Do vendor due diligence. Ask about data security, where consumer data is stored and processed, whether your data trains shared models, subcontractors and audit rights. Our How to Start a Collection Agency guide covers the wider vendor stack.
- Pilot, then monitor. Start with a small, low-balance segment, compare complaint and dispute rates with human-worked accounts, and review samples weekly before scaling.
For the underlying rules, see our FDCPA compliance guide, the Regulation F guide and our compliance overview.
Can AI help price and work a charged-off portfolio?
Yes. Many buyers use statistical and machine-learning models on data-tape fields to estimate liquidation and set bids. The legal exposure is lower than in consumer contact, but model inputs, data permissions and documentation still matter.
Pricing models don’t talk to consumers, so the FDCPA and TCPA issues above mostly don’t apply at the bid stage. What does matter is data. Skip-tracing and consumer-report data used to enrich a file come with permissible-purpose limits, which we cover in our skip tracing guide. And a model is only as good as the tape. Our article on tape signals that move a charged-off bid explains which fields buyers weigh most, and the DebtHub buyer tips collect practical notes for newer buyers.
Once accounts are boarded, the same model often drives work queues: who gets a call, who gets a letter, and who gets a settlement offer. That’s where state automated-decision laws, like Colorado’s version taking effect in 2027, may become relevant, so document what the model uses and why.
What should sellers ask buyers about AI?
Sellers should ask which AI tools touch their former customers, how consent and call caps are enforced, whether automated contact is disclosed, how disputes reach a human, and what records are kept. Clear written answers make buyer approval easier.
Creditors selling charged-off portfolios increasingly include technology questions in buyer due diligence, because a buyer’s conduct can come back to the original brand. Useful questions include:
- Which collection channels use AI, and which vendors provide them?
- How are TCPA consent and revocation tracked for numbers on our file?
- Do AI calls and messages share one Reg F call counter with human collectors?
- Does a consumer always have a path to a live person?
- How long are transcripts and recordings retained, and can we audit them?
- Is our consumer data used to train any shared or third-party model?
DebtMarket, the confidential debt brokerage desk of Triton Financial Solutions, LLC, takes portfolios NDA-first to 200+ qualified buyers, with no public auction, and complete files typically close in 1-14 days. Creditors can start at /sellers/, and qualified buyers can review working files under NDA at /buyers/. More checklists live in DebtHub. We are a marketplace and broker, not a collection agency.
Key takeaways
- No federal law bans AI in debt collection, and none exempts it. The FDCPA, Reg F, TCPA and FTC Act apply to AI the same way they apply to people.
- The FCC’s February 2024 ruling (FCC 24-17) treats AI-generated voices as “artificial” voices under the TCPA, so consent rules apply.
- AI calls, texts and emails must share the same Reg F call caps, time windows and opt-out handling as human contact.
- The CFPB withdrew 67 guidance documents on May 12, 2025, including its algorithm-related adverse action circulars, but the statutes are unchanged and states and private plaintiffs can still act.
- California requires bot disclosure in some online commercial interactions; Colorado’s replacement AI laws take effect January 1, 2027.
- Disclose automation, offer a human, lock sensitive language, keep full records and vet vendors before you scale.
Related reading on DebtMarket
- FDCPA Compliance Guide
- Regulation F Guide
- How to Start a Collection Agency in 2026
- Tape Signals That Move a Charged-Off Bid
Sources (each checked live on October 30, 2026)
- Federal Communications Commission, Declaratory Ruling, Implications of Artificial Intelligence Technologies on Protecting Consumers from Unwanted Robocalls and Robotexts, FCC 24-17, CG Docket No. 23-362 (adopted February 2, 2024; released February 8, 2024).
- CFPB, Regulation F, 12 CFR Part 1006, including § 1006.6, § 1006.14, § 1006.18 and § 1006.100.
- FDCPA, 15 U.S.C. 1692d and 15 U.S.C. 1692e (Cornell LII).
- TCPA, 47 U.S.C. 227 (Cornell LII).
- FTC Act, 15 U.S.C. 45 (Cornell LII).
- Federal Register, Interpretive Rules, Policy Statements, and Advisory Opinions; Withdrawal, 90 FR 20084 (May 12, 2025).
- CFPB, Chatbots in consumer finance (June 6, 2023).
- California Legislative Information, Business and Professions Code § 17941 (SB 1001, operative July 1, 2019).
- Colorado Attorney General, Colorado Anti-Discrimination in AI Law (ADAI) Rulemaking.
- Legal 500, Colorado AI Law In Flux: Comprehensive Replacement Bill Signed After Federal Court Blocks Predecessor’s Enforcement.
This article is general industry information as of the date above. It is not legal advice or an endorsement of any technology vendor.