Supervising GenAI under FINRA Rule 3110: what should you log?

Does FINRA Rule 3110 apply to generative AI?

Yes. FINRA has no rule written specifically for generative AI. Its position is that existing rules, including Rule 3110, already apply. Regulatory Notice 24-09, published June 27, 2024, says FINRA intends its rules and guidance to be technologically neutral and that they "apply when member firms use AI, including Gen AI or similar technologies, in the course of their business."

For supervision, that means Rule 3110. A member firm "must have a reasonably designed supervisory system tailored to its business," and the notice adds that if a firm uses GenAI tools in its supervisory system, its policies and procedures should address technology governance, including model risk management, data privacy and integrity, and the reliability and accuracy of the AI model. The supervisory duty does not change because the work was done by software. What changes is the evidence a principal needs to show the work was reviewed.

What does FINRA say firms should log?

Prompts, outputs, model versions and human review. In the GenAI section of its 2026 Annual Regulatory Oversight Report, FINRA lists monitoring practices that may include "storing prompt and output logs for accountability and troubleshooting," "tracking which model version was used and when," and "validation and human-in-the-loop review of model outputs, including performing regular checks for errors or bias."

These appear in the report as effective practices and considerations, not as new rule text. They still tell you what an examiner is likely to ask about. A firm that cannot say which AI tools are in use, what they were asked, what they produced and who checked the result will have a hard time showing its supervisory system is reasonably designed for that activity.

The same report lists other practices worth adopting: formal review and approval before a GenAI use case goes live, written policies to "develop, implement, use and monitor GenAI," documentation throughout, and testing to understand the capabilities, limitations and performance of the model.

Are AI prompts and outputs books and records?

Some are, depending on content. No SEC or FINRA rule names AI chat logs as a record category. The existing tests still apply. Rule 17a-4(b)(4) requires broker-dealers to preserve communications relating to their business as such, and Rule 3110(b)(4) requires review of incoming and outgoing correspondence. An email to a customer drafted by an AI assistant is correspondence when it is sent, whoever wrote the first draft.

Content standards follow the same logic. Regulatory Notice 24-09 says the content standards of Rule 2210 apply whether a communication is generated by a human or a technology tool. Firms should decide in their written supervisory procedures how they treat prompts, drafts and AI meeting summaries, then confirm the tools can actually retain and produce what the procedures say they keep. Our earlier piece on off-channel communications and AI notetakers covers that retention question in more detail.

How should a firm supervise AI agents?

With tighter controls than chat tools, because agents act. FINRA's 2026 report describes the risk of "AI agents acting autonomously without human validation and approval" and notes that agents "may act beyond the user's actual or intended scope and authority."

The report lists what firms should consider: how to monitor agent system access and data handling, where to place human-in-the-loop oversight, "how to track agent actions and decisions," and how to set guardrails that limit agent behaviors, actions or decisions. In practice, an agent that can send email, update a CRM record or move a file needs its own permissions, its own activity log and a named supervisor. An agent running under a registered person's full account permissions makes it hard to tell later who did what.

What about AI built into vendor products?

It is in scope. Regulatory Notice 24-09 says FINRA rules apply when firms leverage third-party technology, "including through embedded features in existing third-party products." The AI summary in your CRM, the drafting assistant in your email client and the transcription feature in your meeting platform all count.

The 2026 report points firms back to Regulatory Notice 21-29 on supervising outsourced functions. Outsourcing an activity does not outsource the supervisory obligation. Vendor due diligence should now ask whether AI features are on by default, where prompts and outputs are stored, whether customer data is used to train models, and whether logs can be exported for review.

Is FINRA changing how AI-generated communications are reviewed?

It has proposed to. In Regulatory Notice 26-14, published July 9, 2026, FINRA requested comment on modernizing Rule 2210, including replacing the principal pre-use approval requirement for retail communications with a risk-based supervisory standard. The notice acknowledges that applying pre-approval to AI-generated retail communications can be challenging, and states that GenAI tools can be part of a supervisory system if they are vetted, tested and monitored.

The comment period closed September 11, 2026. Until FINRA files and the SEC approves a final rule, current Rule 2210 applies as written. The direction is still useful: FINRA is pointing firms toward supervising AI tools as systems, with testing and monitoring evidence, rather than reviewing every output by hand.

What should you do this quarter?

• Inventory every AI tool and AI feature in use, approved or not, including features embedded in existing vendor products.
• Add a GenAI section to your written supervisory procedures that names approved use cases, prohibited uses and the principal responsible for each.
• For each approved tool, confirm whether it can retain prompts, outputs, model version and user identity, and how long it keeps them.
• Set a human review standard: which outputs need review before use, and what sample of the rest gets checked for errors or bias.
• Give any AI agent its own scoped permissions and an action log, separate from the registered person it assists.
• Update vendor due diligence questions to cover AI features, data use and log export.
• Include AI supervision in your annual Rule 3120 testing so the results support the Rule 3130 CEO certification.

For broader governance, NIST AI 600-1, the generative AI profile of the NIST AI Risk Management Framework, is voluntary guidance that maps generative AI risks to specific actions and can give your policy a recognized structure.

How can s90 help?

s90 helps broker-dealers and dual-registered firms see which AI features are switched on across Microsoft 365 and their other platforms, configure retention and audit logging so prompts and outputs can be produced, and set up scoped access for AI tools and agents. Our agreements include 24-hour evidence production and a 72-hour breach notification commitment. If your written supervisory procedures do not yet cover GenAI, we can help you document what is already in use.