AI Washing: How to Describe Your AI Use in LP Materials Without SEC Risk

1. What AI Washing Is, in Plain Terms

AI washing is overstating your firm's use of artificial intelligence in a way that misleads investors. The word echoes greenwashing, and it describes the same move: dressing up a marketing claim the substance does not support.

For a fund, it usually looks innocent. A deck calls the strategy AI-driven when a couple of analysts use a chatbot for research. A website advertises a proprietary AI model that is really an off-the-shelf tool with a login. A DDQ answer implies AI shapes investment decisions when the investment committee decides the way it always has. None of these start as lies. They start as rounding up, and rounding up is exactly what the SEC has said it will look at.

The fix keeps your AI in the materials. Plenty of firms use AI well and should say so. What changes is the wording: describe what you actually do, in language specific enough to stand behind if someone asks. This guide shows how, starting with why the stakes are higher than they look.

2. Why It Is an SEC Problem, Not a Branding One

AI washing sits inside the Marketing Rule, Rule 206(4)-1, which prohibits false or misleading statements in adviser advertising and requires a reasonable basis for the material claims you make. An AI claim in a pitch deck is an advertising claim, and it carries the same burden as a performance figure.

The SEC has already acted on this. In March 2024 it settled its first two AI-washing cases, against Delphia and Global Predictions, charging that they made false or misleading statements about their use of AI. The firms settled for about $225,000 and $175,000, under the Marketing Rule and the compliance rule (press release 2024-36, sec.gov). The agency's then-chair publicly warned advisers against AI washing around the same time.

The standard the rule sets is a reasonable basis. You are not required to prove your AI is the best on the market or that it drives returns, only to show that what you claimed was true and that you had grounds for it when you made it. That is a lower bar than many firms fear. A specific, modest description of what your tools actually do clears it comfortably, while an inflated one struggles the moment someone asks for proof. Firms that describe their AI plainly rarely have anything to fear from the rule at all.

Worth noting: the SEC did not need a special AI rule to bring those cases. The proposed predictive data analytics rule was formally withdrawn in June 2025, so it is not in force. The tool the SEC used, and will use again, is the ordinary Marketing Rule that has governed adviser advertising for years. That is good news, because it means the standard is familiar: say what is true, and be able to show it.

3. Where AI Washing Hides: PPMs, DDQs, Decks, Websites

The risk is spread across four documents, and it reads differently in each. Knowing where to look is half the job.

The pitch deck is the usual offender. It rewards a punchy line, and AI-driven or AI-powered lands better than a paragraph of accurate detail. The PPM is the opposite risk: it is a legal document, so a loose AI description there carries more weight and more exposure than the same words on a slide. The DDQ is where LPs ask you directly, in writing, and your answer becomes a record you will be held to across the fund's life. And the website is the one everyone forgets, because marketing wrote it, nobody reviews it on a schedule, and it is the first thing an examiner or an LP reads.

The through-line is consistency. If the deck says AI-driven, the PPM says AI-assisted, and the DDQ says you are exploring AI, those three do not agree, and the gap is its own problem. Pick the accurate description once and use it everywhere. The DDQ side, what LPs actually want to hear, is covered in our guide on what to tell LPs about AI.

4. The Phrasing Patterns That Draw Scrutiny

A few phrases do most of the damage. Each one implies a capability an examiner can ask you to prove, and each has a safer, truer version.

The first is the agency swap: language that makes AI the actor. AI-driven, AI-powered, our AI selects, all put the machine in the decision seat. If people make the decisions and AI supports them, say that. The second is the ownership claim: proprietary AI or our model, when the truth is a configured commercial tool. Proprietary means you built it and can document it. The third is the performance tie: any phrase that links AI to returns. Performance claims already carry the heaviest Marketing Rule burden, and attaching AI to them stacks two hard-to-prove claims into one.

The last common one is vague superlatives: cutting-edge, revolutionary, next-generation AI. Some of that is puffery a reasonable investor discounts, but if the claim is material to the strategy an examiner can still test it, and vague words are the hardest to substantiate because they mean nothing specific.

One more deserves its own line, because it is close to what the SEC's first cases turned on: the future tense worn as the present. Describing AI you plan to build as if it already runs is the most direct form of AI washing there is. Aspiration has its place, and a PPM or a DDQ is a statement of current fact, so keep the roadmap separate from the description of what you do today. If a capability is coming, say it is coming. The pattern across all of these is the same: specificity protects you, and inflation exposes you.

5. Before and After: Rewriting the Claims

Here is the pattern applied. The left column is language that draws scrutiny, the middle says why, and the right is a substantiated rewrite you can adapt to your own facts. Use the rewrite only if it is true for your firm; the point is to match the words to the work.

The rewrites share a shape: a specific tool, a specific task, and a clear line where a person takes over. That shape is easy to substantiate because it describes something real, and it reads as more credible to a sophisticated LP than the inflated version ever did.

6. The Substantiation File: Proving What You Say

The Marketing Rule asks for a reasonable basis, so the safe habit is to build one before the claim goes out, not after an examiner asks. That basis is a substantiation file.

The file is short and boring, which is the point. For each AI claim in your materials, hold the evidence that supports it: which tool the claim refers to, what it actually does in your process, who uses it, and where a person reviews its output. If a slide says AI cuts diligence time, the file holds the before-and-after that shows it. If the PPM describes AI-assisted monitoring, the file names the tool and the workflow. The test is simple: could you hand this to a skeptical reader and have them agree the words match the work?

A worked example makes it concrete. Say a slide claims the firm uses AI to review contracts in diligence. The substantiation is a short note: the tool's name, a line on what it extracts (key terms, renewal dates, change-of-control clauses), the analyst who runs it, and one recent deal where it was used, with the reviewer's sign-off attached. A technical dossier is overkill. A paragraph and a pointer to the real workflow is the right size, because the test is only whether a skeptical reader would agree the words match the work.

The file is the same discipline the six-row checklist in our AI governance audit checklist asks of the marketing row, and it earns its keep twice: it holds your materials honest now and hands you the evidence later. Build it as you write the materials, so the claim and its proof stay together, which is exactly the state an examiner wants to find them in.

7. Keep the Record: Books and Records for AI Claims

One more obligation sits behind the claim. Rule 204-2, the books and records rule, requires advisers to keep their advertisements and to retain them for the periods the rule specifies. An AI claim in a deck or on the website is an advertisement, so it, and the substantiation behind it, belongs in the record.

The practical failure here is version drift. A deck says one thing in the spring and a revised deck says another in the fall, and nobody kept the first one. When an LP or an examiner asks what you told investors during a raise, the answer has to be a retained document, not a memory. Keep each version of the AI-facing materials, dated, alongside its substantiation file.

None of this requires new machinery. Your firm already retains advertisements; AI claims just have to ride the same schedule, with their proof attached. The records rule and the substantiation file are two halves of the same habit: say what is true, and keep both the saying and the proof.

8. The Opposite Risk: Saying Nothing at All

There is a mistake in the other direction, and it is getting more common as firms grow cautious. Some managers now scrub AI out of their materials entirely, hoping silence is safer than a claim. It usually is not, because LPs are asking anyway.

DDQs increasingly ask, in writing, whether and how you use AI, and how you protect the data you put through it. A blank or evasive answer reads as either behind the curve or hiding something, and neither helps a raise. The right move is an honest, specific answer: the tools you use, the tasks they support, the point where a person reviews the output, and how you keep confidential data safe.

That last part carries its own accuracy trap, so get it right. If you tell LPs their data and your deal data are protected, describe it truthfully. Commercial plans (Claude Team and Enterprise, the business tiers, and comparable enterprise plans from other providers) do not train on your data, while consumer accounts can unless a user opts out, so confidential material belongs only on the business tiers. Stop short of claiming nothing is ever stored: standard retention still applies to the business chat products, and true zero-retention is a narrower API-level arrangement. Describing your data posture accurately is itself part of not AI washing, because an overstated privacy claim is as substantiable as any other.

9. A Simple Review Step Before Anything Goes to LPs

All of this reduces to one habit: nothing AI-facing reaches an LP without a two-minute check. Build the check into your existing marketing review, so it costs almost nothing.

Three questions clear most claims. First, does this describe what we actually do, or what we wish we did? Second, could we substantiate it today if asked, and where is the proof? Third, does it match what our other materials say? A claim that passes all three is safe to send. A claim that stumbles on any one goes back for a rewrite using the before-and-after pattern above.

Give one person the sign-off, usually the CCO or someone the CCO delegates, and log that the review happened. That log is the same approval record the compliance program expects, so it documents a step you should already run rather than adding a new one. The whole point is to make honest description the default path, not a special effort someone has to remember.

10. Where to Start

Start with an inventory, not a rewrite. Pull every place your firm describes its AI use: the current deck, the PPM, the website, the last few DDQ answers. Read them together in one sitting, which is how you catch the inconsistencies.

Then run each AI claim through the before-and-after pattern and the three review questions. Rewrite what draws scrutiny, delete what you cannot substantiate, and build the substantiation file for what remains. Most firms find the exercise takes a morning and leaves them with materials that are both safer and more credible, because specific beats inflated with the kind of LP who reads carefully.

If you want the wording and the evidence file reviewed before a raise, our SEC Exam-Ready AI Governance package covers the marketing row alongside the rest: it checks your LP-facing AI language, builds the substantiation file, and assembles the governance record an examiner expects, for $17,500. This is operational documentation built alongside your counsel, who owns the legal interpretation. The words you use to describe your AI are worth getting right once, because you will live with them across the whole fund.