SEC 2026 Examination Priorities · Nov 2025
Rosey is a supervisory control for broker-dealers and RIAs. It produces the documented review trail — flagged issues, dispositions, time-stamped reviewer action — that maps directly to your obligations under FINRA 3110 / 2210 and Exchange Act 17a-3 / 17a-4.
Running 10 limited pilots this quarter with SEC/FINRA-registered firms.
Control evidence
Firms hesitate because surfacing AI errors feels like creating a discoverable record of failure. Regulators see it the opposite way: a documented review trail is exactly what a functioning supervisory control looks like.
Rosey's log is not a record of bad AI. It is proof that a human reviewed the output, caught what mattered, and acted on it — retained under 17a-4 and mapped to 3110.
How Rosey maps to your obligations
Regulators are technology-neutral: content and supervision standards apply to AI-generated work exactly as to human work. Below is the mapping we walk compliance teams through in the pilot plan.
| Obligation | What it requires | How Rosey supports it |
|---|---|---|
AI oversight focus, Nov 2025 | Examiners review how firms monitor AI-generated outputs and document human oversight. | Produces an examination-ready record: outputs reviewed, issues flagged, oversight applied — by whom, when, with what disposition. |
Supervision | Firms must supervise all activities regardless of AI involvement; unexplained automated outputs create direct risk. | Adds an auditable review checkpoint over AI-generated documents and an explainable record of what was surfaced, resolved, and by whom. |
Communications with the Public | Content standards apply whether a communication is generated by a human or a technology tool. | Flags errors and unsupported claims in AI-drafted client communications before they go out, supporting principal review. |
Recordkeeping | Firms must retain the records behind their communications and supervision. | Retains a time-stamped log of findings, dispositions, and reviewer adjudication as the supervisory record. |
Sample supervisory log · what Rosey produces
Illustrative sample — not a real firm's data| Document type | Flagged issue | Reviewer | Disposition | Timestamp |
|---|---|---|---|---|
| Client meeting summary | Unsupported performance figure ("up 14% YTD") not present in source transcript | J. Ortega, Principal | Edited before send | 2026-02-11 14:32 ET |
| Quarterly review letter | Forward-looking language lacking required risk qualifier under Rule 2210 | M. Chen, CCO | Returned to advisor | 2026-02-11 15:07 ET |
| Prospect follow-up email | Reference to fund class not offered to retail investors | J. Ortega, Principal | Approved with note | 2026-02-11 15:41 ET |
What the pilot looks like
Fixed fee. Defined scope. Success measured by a delivering a complete, examiner-ready oversight record across a defined set of AI-generated documents.
01
20-minute scoping call
Confirm fit against your AI document workflow and exam calendar. A working conversation about your supervision needs.
02
Mutual NDA
We sign a mutual NDA (template ready) so we can discuss your documents, workflows, and exam calendar in concrete terms.
03
Customized Pilot Plan
We build a pilot plan scoped to your org's AI-generated documents and supervision gaps — deliverables, timeline, and success criteria defined upfront.
04
Pilot decision
If it warrants, a 60-day paid pilot with a defined oversight record as the deliverable — mapped to 3110, 2210, 17a-4.
Q3 PILOT COHORT · LIMITED
Rosey is delivered by the team behind InSummary, working directly with compliance leaders at broker-dealers and RIAs. Pilots are limited this quarter so we can work closely with each firm.
Isn't this just AI supervising AI?
No. Rosey flags issues in AI-generated documents and surfaces them to a human reviewer. Final supervisory sign-off stays with your firm's designated principal. Rosey doesn't approve, reject, or make the supervisory decision — it produces the record that shows a qualified person did.
How your documents are handled
Samples are shared under mutual NDA, stored in an access-controlled environment scoped to the named engagement team, encrypted in transit and at rest, and deleted on request at the end of the pilot. Client data is never used to train models.