Reference · Securities

FINRA Regulatory Notice 24-09: Generative AI Compliance Obligations

Regulatory obligations when using generative AI and large language models

Last reviewed October 11, 2026

At a glance

Issued: 27 June 2024
Title: FINRA Reminds Members of Regulatory Obligations When Using Generative Artificial Intelligence and Large Language Models
Legal effect: Creates no new legal or regulatory requirements and no new interpretations of existing requirements.
Rules named: Rule 3110 (Supervision) and Rule 2210 (Communications with the Public). For the wider set of obligations, the Notice refers to the 2024 Annual Regulatory Oversight Report.
Applies to: FINRA member firms, whether they build a generative AI tool, use a third party's, or use generative AI features embedded in a third-party product.
Official text: Regulatory Notice 24-09 on finra.org

Regulatory Notice 24-09, issued 27 June 2024, reminds FINRA member firms that FINRA's rules continue to apply when they use generative AI and large language models, as they apply to any other technology or tool. It creates no new requirements and no new interpretations of existing ones.Notice 24-09 · Summary FINRA has since built on it in its 2025 Annual Regulatory Oversight Report, in FINRA's 2026 Oversight Report on GenAI and agents, and in a July 2026 proposal to modernize its communications rule.

The rules apply to every generative AI tool a member firm uses

FINRA intends its rules to be technology neutral.Notice 24-09 · Summary They apply whether a firm develops a generative AI tool for its own use or relies on a third party's technology, including generative AI features embedded in third-party products it already runs.Notice 24-09 · Regulatory Obligations Buying the capability from a vendor leaves the obligation with the firm.

Depending on how a firm uses it, generative AI could implicate virtually every area of its regulatory obligations.Notice 24-09 · Regulatory Obligations The Notice discusses supervision and communications directly, and refers firms to FINRA's 2024 Annual Regulatory Oversight Report for a non-exhaustive list of the rest.Notice 24-09 · endnote 6

Evaluate each tool before deployment and supervise it under Rule 3110

Rule 3110 requires each member to maintain a supervisory system, and written supervisory procedures, reasonably designed to achieve compliance with the securities laws and FINRA rules.Rule 3110 · (a), (b)(1) The Notice adds that the system must be tailored to the firm's business. Where a firm uses generative AI inside it, for example to review electronic correspondence, its policies and procedures should address technology governance: model risk management, data privacy and integrity, and the reliability and accuracy of the model.Notice 24-09 · Regulatory Obligations

A member firm should evaluate a generative AI tool before deploying it and confirm that it can continue to comply with the rules that apply, which depend on how the firm deploys the technology.Notice 24-09 · Regulatory Obligations

The 2026 report describes what that looks like in practice: formal review and approval of each use, a supervision, governance or model risk management framework with comprehensive documentation, testing for privacy, integrity, reliability and accuracy, and ongoing monitoring that can include prompt and output logs, a record of which model version was used and when, and human review of outputs.FINRA 2026 Report · Current Practices The entry on the 2026 report covers those practices in full.

Hold AI-generated communications to the Rule 2210 standards

Rule 2210's content standards apply whether a person or a technology tool generates a communication, and FINRA's advertising FAQs address chatbot communications (question B.4) and AI-created communications (question D.8).Notice 24-09 · Regulatory Obligations Every member communication must be based on principles of fair dealing and good faith and be fair and balanced.Rule 2210 · (d)(1)(A) A registered principal must approve each retail communication before the earlier of its use or its filing with FINRA, subject to the exceptions the rule lists.Rule 2210 · (b)(1)

In Regulatory Notice 26-14, issued 9 July 2026, FINRA proposed replacing that pre-use approval requirement with written procedures, appropriate to each firm's business, size and structure, that determine which categories of retail communications need principal approval before use. The proposal cites advances in generative AI among its reasons.Notice 26-14 · Summary It also states that "members are responsible for their communications," whether a human or AI technology generates them.Notice 26-14 · Artificial Intelligence Comments closed on 11 September 2026, and until a change is adopted, Rule 2210 applies as written.

Preserve generative AI records under Rule 4511

Recordkeeping is one of the rules the 2026 report names as implicated by generative AI, alongside supervision, communications and fair dealing.FINRA 2026 Report · Regulatory Obligations Rule 4511 requires members to make and preserve the books and records that FINRA rules and the Exchange Act require, to keep those with no specified retention period for at least six years, and to preserve them in a format and media that complies with SEA Rule 17a-4.Rule 4511

The retention period for a generated record comes from the rule that governs that type of record, as for any other record. The SEC recordkeeping entry covers Rule 17a-4 and Rule 204-2.

Bring vendor generative AI under third-party oversight

The 2025 report took up generative AI inside its third-party risk topic, which cites Rules 3110 and 4370 and Regulation S-P.FINRA 2025 Report · Regulatory Obligations FINRA observed firms proceeding cautiously, generally exploring or implementing vendor-supported generative AI tools to make internal functions more efficient.FINRA 2025 Report · Additional Resources

Among the effective practices it lists: asking prospective vendors whether their products or services incorporate generative AI and, where they do, evaluating the contracts and having them amended as necessary to meet the firm's regulatory obligations, for example with language that prohibits firm or customer sensitive information from being ingested into a vendor's open-source generative AI tool.FINRA 2025 Report · Effective Practices FINRA restated the supervisory obligations that apply when a firm outsources to a vendor in Regulatory Notice 21-29.Notice 21-29

Rule 3110 reaches AI agents

The 2026 report defines AI agents as systems or programs capable of autonomously performing and completing tasks on behalf of a user. It names six risks particular to them (autonomy, scope and authority, auditability and transparency, data sensitivity, domain knowledge, and rewards and reinforcement), on top of the bias, hallucination and privacy risks of generative AI in general.FINRA 2026 Report · Agents

FINRA suggests supervisory processes "specific to the type and scope of the AI agent being implemented," and lists what to consider: how to monitor the agent's system access and data handling, where to place human-in-the-loop oversight, how to track its actions and decisions, and how to set guardrails that limit what it may do.FINRA 2026 Report · Agents Because FINRA's rules apply to the technology a firm uses, an agent's work sits inside the supervisory system Rule 3110 requires. When an agent acts, a supervisory principal should be able to show an examiner:

Who authorized the scope. The approval that set what the agent may do and which systems and tools it may reach, with who granted it and when.
What the agent was given. The instructions, context and data it received for the task, including any customer or other sensitive information.
What the agent did. Each action and decision in sequence, traceable to the task and to the approval behind it.
Where a person reviewed the work. The points at which a person validated or approved the agent's output, and the record of that review.
What was retained. Which of these records are required books and records or business communications under Rule 4511 and SEA Rule 17a-4, and evidence that they were preserved in the required form for the required period.

The 2026 report names auditability as an agent risk because multi-step reasoning can make outcomes hard to trace.FINRA 2026 Report · Agents A supervisory record that runs from the approved scope to the retained result is what makes an agent's work examinable. For the recordkeeping analysis of AI-drafted client messages, see Does generative AI client outreach trigger SEC and FINRA recordkeeping rules?

Control mapping

What a reviewer expects to be able to see.

Obligation	What the system must do	Evidence a reviewer expects
Supervisory system (Rule 3110)	Cover each generative AI use in written supervisory procedures, including any AI inside the supervisory system itself	WSPs naming approved tools, their permitted uses, and the review procedure for each
Evaluation before deployment	Evaluate each tool before it goes live and record who approved it, and for what use	Approval record per tool and use case, with approver, date, and the controls required
Technology governance	Address model risk, data privacy and integrity, and the reliability and accuracy of each model	Approved-model register with versions, test results, and data-handling controls
Communications (Rule 2210)	Route AI-generated communications through content review, and principal approval where the rule requires it	Review and approval records tied to each generated communication
Books and records (Rule 4511)	Retain AI-generated and AI-assisted records under the rule that governs each record type	Retained records in a format that meets SEA Rule 17a-4, retrievable within required timeframes
Vendor oversight	Ask vendors whether their products use generative AI, and set contract limits on firm and customer data	Due diligence files, contract terms, and ongoing monitoring records
Agent oversight	Set each agent's scope, track its actions and decisions, and place human review where the risk calls for it	Scope approvals, action records tied to each task, and review records, retained as books and records where required

Key dates

Timeline, drawn to scale

FINRA guidance on generative AI since Notice 24-09

PublishedAdds AI agents and AI-generated communications

Publication dates of Regulatory Notice 24-09 and the FINRA guidance that builds on it, from finra.org. Status as of October 11, 2026.

10 June 2020FINRA report on artificial intelligence in the securities industry.
13 August 2021Regulatory Notice 21-29 on supervisory obligations when outsourcing to third-party vendors.
9 January 20242024 Annual Regulatory Oversight Report, which Notice 24-09 cites for the wider list of obligations generative AI can implicate.
27 June 2024Regulatory Notice 24-09 issued.
28 January 20252025 Annual Regulatory Oversight Report addresses generative AI within third-party risk.
9 December 20252026 Annual Regulatory Oversight Report gives generative AI its own topic, including AI agents.
9 July 2026Regulatory Notice 26-14 proposes modernizing Rule 2210, citing generative AI.
11 September 2026Comment period on Regulatory Notice 26-14 closed.

Sources cited

Common gaps

Where a firm's generative AI adoption most often runs ahead of its supervisory system.

Shadow adoption. Staff use tools that were never approved, inventoried, or covered by written supervisory procedures. A firm cannot supervise a tool it does not know it runs.
Procedures written for deterministic systems. WSPs that assume the same input always gives the same output leave generated content unaddressed. Procedures need to say how output is reviewed, as well as who approved the tool.
No record of what the model was given. Supervision and recordkeeping both assume an output can be traced to its inputs. Prompts and context that are not retained break that trace.
Vendor attestations in place of supervision. The supervisory obligations in Notice 21-29 apply to outsourced activity.Notice 21-29 A vendor's statement about its own controls leaves the firm with nothing to show for its own supervision.
Communications that skip review. AI-assisted drafts that reach customers without content review are usually a routing problem: the draft leaves the tool through a channel the review workflow does not cover.
Retention that misses AI output. Chat transcripts, generated summaries, and model output inside customer communications often sit outside the retention system.
Agent actions outside books and records. An agent that sends a message, updates a record, or calls another system on a person's behalf produces activity the firm must supervise and, where it is a required record or business communication, preserve. Action logs kept only inside the agent platform often sit outside the retention system, and the approval that set the agent's scope is often not kept at all.

In practice

FINRA 24-09: evidence for supervision and recordkeeping →

Last reviewed October 11, 2026. This reference summarises publicly available regulatory guidance and is provided for general information. It is not legal advice. Obligations depend on an institution's charter, registration status, size, and activities. Verify against the primary sources cited above and consult counsel before relying on any summary here.