SEC Marketing Rule LLM Disclosure: What RIAs Must Disclose and What They Must Prove

What the Marketing Rule says about AI, and what it does not

The Marketing Rule applies to any investment adviser registered, or required to be registered, with the SEC under section 203 of the Advisers Act that directly or indirectly disseminates an advertisement. It contains seven general prohibitions on activity that could be false or misleading, and these apply to all advertisements. It permits performance information only when specific conditions are met. It prohibits testimonials and endorsements unless disclosure, oversight and disqualification provisions are satisfied, and it prohibits gross performance unless net performance is also presented. The rule took effect May 4, 2021, together with amendments to Rule 204-2 and Form ADV.

None of those provisions names AI or LLMs, and the rule text, the adopting materials and the Division of Examinations' marketing rule risk alerts reviewed for this guide contain no language requiring an adviser to say that AI was used to create an advertisement. Counsel should confirm that reading against the full rule text and current staff guidance, including the staff FAQs, which were updated in January 2026. As of October 2026, the defensible position is that the rule is tool-neutral.

Tool-neutral does not mean AI-neutral in practice. The standards are content- and outcome-based, so an LLM that produces an unsupported performance claim creates the same violation a person would. The difference is volume and speed: a model can generate and adapt many variants of content before a reviewer sees any of them.

Obligations that apply to AI-assisted content regardless of disclosure

Four obligations matter most for LLM-assisted workflows.

First, the general prohibitions on false or misleading content. A fluent draft is not evidence of accuracy, and the adviser, not the model, is the party that disseminates the advertisement.

Second, substantiation. Division of Examinations marketing rule materials say staff review whether advisers have a reasonable basis for believing they can substantiate material statements of fact. A model has no inherent way to show where a statement came from, so the workflow must supply that support.

Third, performance and testimonial conditions. Staff also review whether performance advertising, including extracted and hypothetical performance, complies with the rule. Net and gross presentation, and the disclosures that must be presented clearly and prominently, are exactly where generated text can drift from approved language.

Fourth, books and records. Amended Rule 204-2 includes the recordkeeping requirements tied to the Marketing Rule, and Form ADV was amended to collect more information about marketing practices. Specific retention periods and required record types should be taken from the rule text and set deliberately for AI-assisted content.

Adjacent disclosure duties that LLM use may implicate

Separate from the Marketing Rule, two disclosure duties may bear on LLM use. Form ADV Part 2 requires delivery of a brochure to clients and prospective clients, and its instructions state that advisers are fiduciaries who must make full disclosure of all material facts relating to the advisory relationship and seek to avoid conflicts of interest. Whether a particular use of an LLM creates a material fact or conflict that belongs in the brochure or in client communications is a judgment for the firm and its counsel. The duty is stated in general terms and is not applied to AI in the material reviewed here.

The compliance program rule, Rule 206(4)-7, is the more direct hook. The Division of Examinations risk alert on marketing-rule compliance says advisers should consider whether to update written policies and procedures, and it quotes the adopting release's expectation of "objective and testable means reasonably designed to prevent violations of the final rule." Staff review whether such policies are adopted and implemented. A policy that says nothing about LLM-assisted drafting is hard to call implemented for that activity.

The SEC's July 26, 2023 proposal on predictive data analytics conflicts of interest is worth monitoring. As proposed, it would require written policies and procedures, an annual review and recordkeeping, and would not allow disclosure alone to address the conflicts. An SEC advisory committee draft recommendation describes the proposal's technologies as including large language models. This page does not treat the proposal as in force.

What examiners are likely to ask for, and why review after the fact falls short

The examiner-relevant objects in the provided guidance are the advertisement as published, the support for material statements of fact, performance data and disclosures, and the written policies and procedures. Interaction-level records of prompts, outputs and approvals are not named as a requirement. They are a design choice, and a useful one: they let a firm show what content was permitted, what the model produced and what controls applied, rather than asserting that a review happened.

The "objective and testable" language points toward controls that can be verified, not only a reviewer's checklist. A post-publication sample review can find problems, but it cannot show that a given advertisement was subject to an enforced rule when it was drafted. Enforcement in the execution path can act before content leaves the workflow. That is a design option rather than a mandate, and it carries tradeoffs: policies must be written precisely enough to enforce, escalation paths need staffing, and false blocks will slow drafters until rules are tuned.

For each AI-assisted advertisement, a firm should be able to answer five questions: what was requested, which policy applied, what action was taken, what the model returned and who approved the result.

Where Trussed fits

Trussed is control, provenance and evidence infrastructure for consequential AI. It enforces policy in the execution path across models, agents, tools and MCP servers: policy is enforced at runtime, not reviewed after the fact. Actions are allow, redact, block, rate limit or escalate for human approval, applied at the level of individual interactions and tool calls. That scope matches the control points above for prompt inputs, generated output and pre-publication approval.

Enforcement generates audit-ready evidence continuously. Trussed deploys self-managed in the customer's cloud, with configurable retention and data residency, so a firm can set retention against its own books-and-records analysis. Trussed has run continuously in production in a regulated enterprise for more than 18 months, at 99.95% uptime on live production traffic, with more than 100 models under one policy plane and more than 30 GB of audit evidence generated per day. Its customers are in insurance, pharma, healthcare and education, and insurance is where it is deepest.

Trussed does not decide what the Marketing Rule requires, and it does not replace a firm's compliance judgment or counsel's reading of its disclosure duties. Transaction-level governance, model lineage and end-to-end tracing of agentic workflows are not part of what is described here.