Governance as a Runtime Control Problem
Dealerships are introducing AI agents into F&I workflows to assist with product recommendations, rate presentations, document generation, and regulatory disclosures. These are not experimental tools; they interact directly with credit terms, customer financial data, and disclosure content that is already subject to federal lending and consumer protection law. The governance question for compliance leaders is not whether AI should be used in F&I, but whether the agent's actions can be constrained, attributed, and reconstructed after the fact. That framing points toward runtime controls: mechanisms that operate while the agent is executing a task, rather than static policy statements that describe intended behavior without enforcing it.
Regulatory Obligations That Already Apply to Agent Behavior
No AI-specific statute currently governs dealership AI agents. Instead, existing rules extend to them by virtue of the function they perform. TILA and Regulation Z require accurate disclosure of APR, finance charges, and payment schedules in consumer credit transactions, an obligation that attaches to the disclosure itself, not to who or what generated it. ECOA and Regulation B prohibit discrimination in credit transactions and apply directly to automated or algorithmic tools used in pricing or product recommendations. The CFPB has stated in Circular 2023-03 that creditors using complex algorithms or AI models for credit decisions must still provide specific, accurate adverse action reasons; model complexity is not a defense for vague or inaccurate notices. The FTC has separately stated that companies deploying AI tools remain responsible for ensuring outputs are accurate and non-deceptive under existing consumer protection law. Compliance teams should treat these as binding constraints on agent design, not as background context.
Building Audit Trails That Satisfy Examiner Review
Recordkeeping expectations under TILA and related consumer finance law generally require institutions to retain sufficient records to reconstruct individual credit transactions, including the disclosures presented and terms offered. Applying this to AI agents means logs must capture more than a final output. An auditable trail should record the agent's identity, the specific action it invoked, the input parameters it used, and the resulting output artifact delivered to the customer. This level of detail allows a compliance team or examiner to reconstruct the agent's decision path for a specific transaction rather than relying on a general description of how the system is supposed to behave. Retention periods for these logs should align with existing dealership recordkeeping practices under federal and state lending rules, which vary by jurisdiction and should be confirmed separately.
Common Questions on F&I Agent Governance
Does agent governance replace an existing compliance program?
No. Agent governance implements controls that support existing obligations under TILA, ECOA, and the FTC Safeguards Rule. It does not create new regulatory requirements; it enforces the existing ones at the point where an agent acts.
Why does an agent need its own identity separate from the associate?
A distinct agent identity allows actions to be attributed precisely during audit review. Without it, logs cannot reliably distinguish which outputs originated from a human decision versus an automated one.
How does fair lending risk apply to AI agent recommendations?
ECOA and Regulation B apply to any tool used in credit decisioning or pricing, including AI-driven recommendations. Governance should include testing agent outputs for disparate impact consistent with fair lending requirements.