Reg FD was written in 2000. It governs selective disclosure to humans. It does not govern selective retrieval inside enterprise AI tools now used by institutional investors to prep for every meeting on the calendar.
That is the regulatory gap. It is large. It is widening. The first enforcement action or shareholder lawsuit built around it will define how the SEC treats the AI Disclosure Layer for the rest of the decade.
What Reg FD covers, and what it doesn't
Reg FD requires that material non-public information disclosed to a market professional be simultaneously or promptly disclosed to the public. The regulation assumes the channel is a human conversation. It assumes the audience is identifiable. It assumes the disclosure either happened or it didn't.
The 2026 operating reality is different. A portfolio manager at a major institutional fund opens ChatGPT Enterprise. Asks for a summary of an issuer. The summary draws on a private earnings-call transcript ingested through a vendor agreement, a Reddit thread tagged to an old executive comment, and an internal-newsletter blurb scraped from a partner site. The PM uses that summary to prepare for a one-on-one with the CFO.
The summary is, in part, sourced from material the company never intended to be retrievable in that combination. Nothing in Reg FD addresses the combinatorial output of an AI retrieval system. The regulation was written before the surface existed.
Three scenarios the current framework does not cover
- Selective retrieval. ChatGPT Enterprise returns one answer for paying enterprise customers and a blander answer for free users. The enterprise answer carries higher-quality retrieval. The free answer does not. Is that selective disclosure? The regulatory text doesn't reach the question.
- Fair-disclosure timing. An issuer makes a material change in guidance. The 8-K is filed at market close. The AI engines, drawing on transcript pipelines that update on a lag, continue returning the prior guidance to PMs running queries for the next 48 hours. The 8-K is filed. The Machine Narrative has not updated. The regulatory framework has no view on the gap.
- Hallucinated material information. A model fabricates a litigation settlement, an executive departure, or a guidance number. The hallucination enters the Investor Retrieval Surface and influences trading. The issuer did not create the information. The engine did. No party in the existing disclosure framework is currently accountable for the trade.
Where the first enforcement action comes from
Three vectors are live.
The SEC's Office of Enforcement will pursue an issuer for the first time when AI Summary Risk intersects with an obvious materiality threshold — most likely during a deteriorating quarter, when stale AI-summary guidance contradicts updated filings.
Plaintiff-side securities litigation will build the first case around Retrieval Distortion and 10b-5. The path of least resistance: a hallucinated negative claim that depressed the stock, retained by the engine for weeks, traceable to a fabrication.
State attorneys general will move faster than the SEC. New York and California already have AI-specific disclosure rules in early-draft form. The first state-level enforcement action against an AI vendor — not the issuer — will reshape who carries the risk.
The 12-month watch list
Four indicators that signal the regulatory framework is about to break against the issuer that didn't prepare.
- The first SEC Division of Enforcement comment letter that references an AI tool by name in connection with an investor-disclosure inquiry.
- The first plaintiff-side complaint that pleads Retrieval Distortion as a 10b-5 theory.
- The first state AG action against an AI vendor — not an issuer — for misrepresenting materially relevant information about a public company.
- The first earnings call where an analyst opens by quoting an AI summary back to the CFO and asking them to confirm or correct it on the record. That is already happening informally on calls. It will be the formal moment the issue goes public.
Why the SEC is behind
The agency is structurally configured to regulate disclosure into known channels. The AI Disclosure Layer is not a known channel. It is a synthesizing layer above all of them. Regulating it requires a vocabulary the SEC does not yet have, a technical staff it does not yet employ, and an enforcement theory that has not been litigated.
The agency will catch up. The cost of the catch-up will be measured in misframed retrievals, eroded valuations, and shareholder actions that build the case law one painful settlement at a time.
The general counsel community will move first. Then the corporate bar. Then the SEC will write the rule that codifies what the bar already figured out. That is the order regulation always travels in. The issuers that build Retrieval Governance discipline now will be the ones counsel uses as the safe-harbor template when the rule finally arrives.
FAQ
What is Reg FD and why doesn't it cover AI summaries?
Reg FD (Regulation Fair Disclosure), adopted by the SEC in 2000, requires issuers to disclose material non-public information broadly rather than selectively to market professionals. It was written for human-to-human disclosure channels. It does not address AI retrieval systems that combine multiple sources into a single summary delivered to a portfolio manager.
What is the AI Disclosure Layer?
The synthesizing layer above all corporate disclosure channels — filings, press releases, earnings calls, executive interviews — where AI engines ingest, combine, and surface materially relevant information to institutional investors. It is not a known channel the SEC has built a regulatory framework around.
What is Retrieval Distortion?
When an AI retrieval system returns a materially incorrect summary about an issuer — through hallucination, stale data, or combinatorial error — and that distorted summary influences investor behavior. The first 10b-5 case built around Retrieval Distortion will define the cause of action.
Who will face the first enforcement action — the issuer, the AI vendor, or both?
Three vectors are live: the SEC's Office of Enforcement (most likely against an issuer during a deteriorating quarter), plaintiff-side securities litigation (against issuers under 10b-5), and state attorneys general (likely against AI vendors first, in New York or California).
What can a general counsel do today to prepare?
Build Retrieval Governance discipline — the issuer-side workflow for monitoring how AI engines describe the company, identifying material gaps between the AI summary and the filed record, and documenting correction efforts. The issuers that build this now will be the ones counsel uses as the safe-harbor template when the rule arrives.
