AI outreach compliance for mortgage.

The pattern a bank will sign off on.

Where autonomous AI voice fails the TCPA bar.

Outbound AI voice calling to a consumer mobile without prior express written consent is a TCPA risk regardless of whether the voice is AI-generated or human. The FCC's 2024 ruling on AI-generated voices in robocalls brought clarity (AI voice falls under the existing TCPA framework) rather than exemption. Enterprise AI voice products that pace toward 100M-plus outbound calls per year are acting at a volume where even a low per-call violation rate produces nine-figure exposure.

The compliance-safe posture for a loan officer is to treat AI voice as a human-triggered tool (the LO clicks "call," the AI handles the dial and the opening, the LO steps in for the conversation) rather than as an autonomous outbound dialer. LO Radar does not operate outbound AI voice.

Where AI credit pulls trigger FCRA CRA status.

An AI system that routinely pulls consumer credit, aggregates it across a book of past borrowers, and sells the output back to the loan officer as a product becomes a consumer reporting agency under FCRA and inherits the full CRA obligation set: dispute handling, permissible-purpose verification, accuracy standards, consumer-facing access rights. Mortgage AI vendors that have not explicitly structured themselves to avoid CRA status carry hidden FCRA exposure. LO Radar is not a CRA; credit-improvement flags come from FCRA-clean soft-pull tier-band aggregations only, documented in the Credit Signal Methodology whitepaper.

Where LLM hallucinations become TILA exposure.

A free-form LLM asked to draft a refi pitch to a borrower can produce a plausible-sounding but fabricated rate, APR, program term, or disclosure. In a borrower-facing message, a fabricated figure is a TILA and state UDAAP issue regardless of the technology that generated it. The compliant pattern is to render numeric values from structured sources (a real rate feed, the real LOS record, a real program-rules table) and to flag any draft where the LLM cannot cite a structured source for every numeric claim. LO Radar renders rate figures from the Freddie Mac PMMS feed and loan terms from the imported LOS record, and refuses to generate a numeric value that cannot be traced to a structured source.

What the audit log must include.

For a bank compliance review to accept an AI outreach tool, the audit log needs:

• The draft version shown to the loan officer, with every revision.
• The loan officer's edits.
• The approver's identity and timestamp.
• The sending channel used.
• The compliance flags that fired: RESPA posture, TILA rate basis, TCPA consent status, state disclosure, licensing gate on the LO's state coverage.
• The outcome: sent, deferred, abandoned.
• Attribution tokens so a closed loan can be traced back to the specific outreach that produced it.

LO Radar logs all seven on every outreach. The Branch Compliance Dashboard surfaces the log to the branch compliance officer in a read-only view, filterable by loan officer, trigger, and date. See /security for the full set.