# AI disclosure

Our collection calls and email are handled by an AI agent. We tell you that. Here is the policy and the legal context.

## What our agent says

Every call our voice agent makes opens with an AI disclosure as part of the mandatory first 10 seconds (defined in `config/prompts/voice/preserve_opening.md`):

> "Hello, this is \[agent name], an AI assistant calling on behalf of \[creditor name]." "This call is being recorded."

In states with an extra AI-disclosure rule, the engine inserts a more explicit AI-disclosure line (`ai_disclosure_line`) on top of the opening.

If you ask at any point in the call "are you a real person?", the agent must say:

> "No, I'm an AI assistant calling on behalf of \[creditor name]."

We never claim to be human. This is a strict rule, enforced by post-call review.

## Which states require AI disclosure

These are the values currently set in `config/compliance/states.json`:

| State               | `ai_disclosure_required`                      | Deadline          | Notes                                                                                                                                                     |
| ------------------- | --------------------------------------------- | ----------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------- |
| California          | Yes                                           | Within 10 seconds | AB 2905 ($500/violation)                                                                                                                                  |
| Texas               | Yes                                           | Within 30 seconds | SB 140 ($1K–$10K/violation)                                                                                                                               |
| Colorado            | Yes (stayed)                                  | n/a               | CO AI Act stayed by federal magistrate Apr 28 2026 — our engine still treats disclosure as required as a conservative default, but enforcement is on hold |
| Utah                | No (state-level rule applies on request only) | n/a               | Utah AI Policy Act requires disclosure only when asked. The Utah row in our config has `ai_disclosure_required: false`                                    |
| All other 47 states | No                                            | —                 | Our universal opening already discloses                                                                                                                   |

For the full per-state matrix, see `config/compliance/states.json`.

## Our policy beyond legal requirements

The opening AI disclosure runs on every call, regardless of state requirements. Reasons:

* **It's the right thing to do.** People deserve to know who or what they're talking to.
* **It builds trust.** Trying to hide that the caller is AI undermines the credibility of the whole conversation.
* **It reduces complications.** Honest disclosure prevents later disputes about "I thought I was talking to a person".

## You can ask the agent to switch to a human

If you'd prefer to talk to a human, tell our agent. The agent can record a callback request via the `capture_callback` tool. Scheduling a callback within compliance rules is built in; a human handoff workflow with same-day response SLA is planned but not yet available.

## What the AI can and can't do

Our agent can:

* State the invoice number, amount, and due date.
* Take payment authorizations (subject to the creditor's authorized terms).
* Receive disputes via the `capture_dispute` tool.
* Honor cease-of-contact requests via the `capture_cease` tool (scope `call`, `email`, or `all`).
* Record attorney representation via `capture_attorney`.
* Record bankruptcy disclosures via `capture_bankruptcy`.
* Record identity-theft assertions via `capture_identity_theft`.

Our agent can't:

* Negotiate outside the bounds the creditor has authorized.
* Imply attorney, court, or government affiliation (FDCPA §807).
* Invent urgency or scarcity that isn't materially true.
* Cite specific other debtors as social proof.

## Email AI disclosure

Our outbound emails are sent under the creditor's brand and are templated. Inbound replies are processed by an LLM-based classifier that buckets each reply into one of six intents: `dispute`, `pay_intent`, `callback_request`, `unsubscribe`, `cease_all_contact`, or `other`. Destructive intents (dispute, unsubscribe, cease\_all\_contact) are gated on DKIM-pass + sender-on-file checks to prevent spoofed replies from taking action.

## What about the LLM that processes call recordings

After each call, a post-call LLM pass extracts structured outcomes from the transcript (payment commitments, dispute signals, compliance signals). This is automated processing of conversation data. It doesn't change anything about the call itself.

***

Last reviewed: 2026-05-12 by Compliance Lead. **TODO: external counsel review.**


---

# Agent Instructions: Querying This Documentation

If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter:

```
GET https://help.moderncollections.io/for-debtors/compliance-and-rights/ai-disclosure.md?ask=<question>
```

The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.