FINRA Rule 3110 requires a broker-dealer to establish and maintain a supervisory system reasonably designed to achieve compliance with securities laws and FINRA rules. It covers written supervisory procedures, designated supervisors with defined responsibilities, review of correspondence and internal communications, and internal inspections. It is the rule under which supervisory failures are most often charged, because it asks not whether misconduct occurred but whether the firm had a system that should have caught it.
AI reaches the rule through conduct rather than through technology. Where a model drafts client correspondence, summarizes a research note, screens a communication or takes an action inside a workflow, that output is the firm's, and the question is whether the firm supervised it. "Reasonably designed" is assessed on what the firm could show, which means the supervisory record has to exist before anyone asks for it.
How it differs from model risk guidance. Model risk guidance asks whether the model is fit for purpose. Rule 3110 asks whether the firm supervised the conduct the model participated in. The second question survives the first: a validated model used without supervision is still a 3110 problem.
Why agents raise the stakes. A supervisory system built for people assumes a reviewable artifact — a message sent, a document produced. An agent may take twenty actions between two artifacts. If the intermediate steps are not recorded, the supervisory record has a hole in exactly the place an examiner will look.
Smartflow's Regulatory Examination Suite generates examination evidence packages via API.
What a supervision record has to contain
The gap most firms discover late is between logging and supervision evidence. A log shows that something happened. A supervision record shows that a system designed to catch it was in force at the time, who was accountable, and what the review found. Assembling the second from the first, after an examination request arrives, is where most of the cost of an examination sits.
Producing it from runtime records means three things have to be captured at the moment of the call rather than reconstructed: the identity of the actor, including when the actor is an agent; the policy version in force; and a record that cannot be edited afterwards. Those are covered in tamper-evident audit logs, AIDA agent identity and the examination readiness suite, with sample examination reports showing the output. The wider obligations for broker-dealers sit alongside model risk guidance — see SR 11-7 and SR 26-2 — and the sector view is on the financial services page.
Related terms
SR 11-7 and SR 26-2 · Information barrier · NIST AI RMF · FFIEC · AI incident response · Full glossary
Put this in the path of your own agents.
Policy enforced inline between your agents and every model and tool they reach, with a record bound to the human who owns it.
Request a Demo Read the docs