Compliance Guide
AI Ad Targeting Fair Lending Compliance: Exam Scrutiny and Runtime Governance Controls
Treat any AI model or agent that influences who sees, receives, or is offered a credit or credit-adjacent product as potentially in scope for fair lending review. Legal and compliance should make the final scope decision against current law and supervisory expectations.
Runtime governance controls, including policy enforcement, scoped agent permissions, and tool-call logging, can produce evidence of how targeting was constrained and monitored. They do not establish compliance on their own.
Where oversight evidence comes from
Scope: influence over credit marketing is the working test
- Scope decision
- Documented determination of which marketing models and agents are in scope, and who made it.
Where risk arises in an AI-driven marketing pipeline
- Targeting constraints
- Restricted attributes and proxies enforced before campaigns are configured.
Documentation, testing, and monitoring
- Delivery testing
- Periodic disparity review of audience delivery, with retained results.
Runtime governance controls that produce evidence
- Runtime audit trail
- Logged agent identity, permissions, tool calls, and policy decisions.
Evidence to be able to produce on request
- Model and agent inventory with documented fair lending scope determinations and rationale
- Targeting logic documentation, including restricted attributes, proxy review, and seed audience definitions
- Policy rules applied to credit campaign parameters, with version history
- Agent identities, permission scopes, and approval records for sensitive changes
- Tool-call logs showing submitted parameters, policy decisions, and approvers
- Disparity testing results, escalation records, and retention schedule
Review your evidence gaps
If AI agents touch audience selection or campaign configuration, a review of runtime permissions, policy enforcement, and logging can show where your oversight evidence is thin.
Talk to an Expert