← All news
BreakthroughMAY 28, 2026 · RIA · SEC ENFORCEMENT

SEC Marketing Rule + AI-Washing: 3 Mistakes That Got RIAs Fined

Three SEC AI-washing orders — Delphia, Global Predictions, Rimar Capital — show the same pattern. The AI wasn't the problem. The documentation was.

By Kadin Nestler · May 28, 2026 · 11 min read
Share X LinkedIn Email
AI-washing enforcement: documented cases
  1. 1
    Delphia (USA) Inc.
    Mar 18, 2024 · IA Release 6573, AP 3-21894 · false claims about AI/ML use of client data in investment process
    $225K
  2. 2
    Global Predictions Inc.
    Mar 18, 2024 · IA Release 6574, AP 3-21895 · "expert AI-driven forecasts" + "first regulated AI financial advisor" without support
    $175K
  3. 3
    Rimar Capital USA / Itai Liptz
    Oct 10, 2024 · Release 2024-167 · $460K from CEO + $60K from board member + 5-year industry ban · fake AI automated trading
    $520K+

The Securities and Exchange Commission has brought three documented AI-washing enforcement actions against investment advisers since March 2024. Total monetary sanctions: north of $860,000. One five-year industry ban. Zero of the three cases turned on whether the AI worked. All three turned on what the adviser wrote down — and what the adviser couldn't produce when the SEC Division of Examinations asked for the file.

This is the pattern across Delphia, Global Predictions, and Rimar Capital. Read the three orders side by side, the same three mistakes show up. Each one is mechanically preventable with a pre-publish review workflow that takes roughly fifteen minutes per marketing piece.

What follows is documentation of the enforcement record, not legal advice. The orders are public. The civil-penalty math is public. The disclaimers the firms didn't include are reconstructable from the orders themselves.

The three orders, in plain language

Delphia (USA) Inc. — Investment Advisers Act Release No. 6573, AP File No. 3-21894, March 18, 2024. From 2019 through 2023, Delphia represented in Form ADV filings, on its website, and in a press release that the firm used machine learning to incorporate client data into its investment process. The SEC order found that the underlying ML pipeline either did not exist or did not function as represented. Settled for a $225,000 civil penalty, censure, and cease-and-desist. Violations cited: Section 206(2) and 206(4) of the Investment Advisers Act of 1940, Rule 206(4)-1 (the Marketing Rule), and Rule 206(4)-7 (the Compliance Rule).

Global Predictions Inc. — Investment Advisers Act Release No. 6574, AP File No. 3-21895, March 18, 2024. The firm's public website claimed its technology incorporated "expert AI-driven forecasts" and described the firm as the "first regulated AI financial advisor." The SEC order found neither claim was substantiated at the time the statements were made. Settled for a $175,000 civil penalty, censure, and cease-and-desist. Same statutory violations as Delphia.

Rimar Capital USA, Inc. / Rimar Capital, LLC / Itai Liptz / Clifford Boro — SEC Press Release 2024-167, October 10, 2024. Liptz, the CEO, raised approximately $4 million from investors after telling them the Rimar entities used artificial intelligence to perform automated trading across equities, futures, and crypto. The SEC order found the AI automated-trading system did not exist as described. Liptz agreed to a total of more than $460,000 in disgorgement, prejudgment interest, and civil penalty, plus a five-year associational bar. Board member Boro paid a separate $60,000 civil penalty. This one was charged as fraud — Section 17(a) of the Securities Act of 1933 and Section 10(b) of the Securities Exchange Act of 1934 — not just Marketing Rule.

Three different fact patterns. One identical compliance failure mode underneath.

Mistake 1: A materially misleading claim about AI capability

The first mistake is the obvious one and also the easiest to repeat. An adviser writes copy that describes what the firm wants the AI to do — what the roadmap says it will do — rather than what the system, today, actually does.

Delphia's website language about machine learning incorporating client data was aspirational copy. The infrastructure to actually do that, at the scale represented, was not in place during the period the statements were live. Global Predictions described the firm as the "first regulated AI financial advisor" — a superlative claim with no supporting evidence the SEC could find. Rimar described a fully automated AI trading system that, per the SEC order, was not operational.

Rule 206(4)-1(a)(1) — the prohibition on untrue statements of material fact in adviser advertisements — does not care about intent. It cares about whether the statement was true at the time it was made, and whether the adviser can produce evidence to substantiate it.

THE ELEVEN-WORD DISCLAIMER THAT WOULD HAVE SAVED DELPHIA
"Machine-learning techniques are used in select components of our investment research." That sentence, swapped in for the broader "we use AI/ML to incorporate client data into our investment process," would have moved the claim from materially overstated to defensibly narrow. None of the three firms had to abandon the AI story. They had to right-size it to the working surface area.

Mistake 2: No contemporaneous documentation of when and how AI was used

This is the mistake the SEC orders punish hardest, because it converts a possibly fixable disclosure problem into a Compliance Rule violation. Rule 206(4)-7 requires advisers to adopt and implement written policies and procedures reasonably designed to prevent violations of the Advisers Act. When the Division of Examinations sends the document request for "all materials substantiating the use of AI/ML as described in your website and Form ADV," the firm has to produce a file.

Delphia could not. Global Predictions could not. Rimar's claims were fabricated outright and there was nothing to produce.

A contemporaneous AI-usage file is unglamorous. It is: which model, which version, which date the model was put into production, which date it was retired or replaced, which workflows used it, what percentage of decisions in those workflows were materially influenced by the model's output, what human review was applied. A short memo per model per quarter. Five pages, signed by the CCO.

The absence of this file is what turns a one-line marketing claim into a Rule 206(4)-7 violation. The presence of it — even a sloppy version — is what gives the adviser standing to argue the claim was accurate at the time.

Mistake 3: Hypothetical or projected AI performance without Marketing Rule disclaimers

The third mistake is more specific to advisers running model-driven strategies. Rule 206(4)-1(d) imposes specific requirements on the use of hypothetical performance — including model performance, projected performance, and backtested performance — in advertisements. The adviser must have policies and procedures reasonably designed to ensure that the hypothetical performance is relevant to the likely financial situation and investment objectives of the intended audience.

In practice, this means a publicly accessible website cannot show "what our AI model would have returned" without a substantial wall of disclaimers and an audience-relevance analysis. The Marketing Rule FAQ updated by the SEC staff has been clear on this: hypothetical performance on a general-access website is presumptively non-compliant because the adviser cannot tailor the presentation to a specific intended audience.

The pattern the orders share

Read the three orders sequentially and the same compliance gap appears:

  • Marketing or IR copy was drafted by a non-compliance owner, often the founder or head of growth.
  • The copy described AI capability in terms broader than the engineering reality.
  • No pre-publish review against an actual evidence file was conducted.
  • The Marketing Rule advertisement went live across the website, Form ADV, press releases, and pitch decks.
  • Examination staff or enforcement staff requested substantiation. The substantiation file did not exist.
  • The cost stopped being a copy-edit and started being a civil penalty, a censure, and — in Rimar's case — a five-year bar.

Cost of step 1 through 4: zero, except the founder's time. Cost of step 5 through 6: $175,000 minimum on the documented record. $520,000 plus a career-ending bar at the top end.

What a pre-publish workflow actually looks like

The fix is not exotic. It is the same workflow that fixed the equivalent problem two product cycles ago for ESG marketing claims and one product cycle ago for ETF tracking-error claims.

Step one: AI claim inventory. Pull every place the firm describes its use of AI or ML. Website, Form ADV Part 2A, Part 2B brochure supplements, ADV Part 3 (Form CRS), press releases for the trailing twenty-four months, pitch decks in active circulation, LinkedIn company-page copy, podcast appearance descriptions, conference panel bios. Most firms find between fifteen and forty surfaces.

Step two: Diff against actual practice. For each claim, write one sentence describing the literal current-state system that backs it. If the sentence is harder to write than the marketing copy, the marketing copy is overstated.

Step three: Build the evidence file. Per model, per quarter: a short memo with model identifier, version, in-production date, retirement date if applicable, workflows touched, percentage of decisions materially influenced, human-review checkpoint. Signed by CCO.

Step four: Pre-publish review. Any new marketing piece referencing AI gets one compliance review against the evidence file before it goes live. Fifteen minutes per piece.

Step five: Quarterly attestation. CCO signs off quarterly that the website and Form ADV language still match the evidence file. The signature is the artifact.

The /ria/marketing-rule-preflight tool runs steps one, two, and four against an uploaded marketing piece in roughly ninety seconds. It is not a substitute for CCO judgment. It is a substitute for the CCO catching every overstated adjective by hand at midnight before the website push.

Cost framing

Civil penalty floor across the three documented orders: $175,000. Median: $225,000. Top end with disgorgement and an industry bar: north of $520,000.

Reputational cost — the part the orders don't itemize — is the harder line item. Each of these firms now appears in every SEC AI-enforcement summary written for the next decade. Every prospect who runs even shallow diligence on the firm name sees the press release. The civil-penalty money is the small number on the invoice.

A pre-publish review tool that runs against the marketing piece before it ships costs the firm somewhere between the cost of a single hour of outside counsel time and the cost of a junior compliance analyst's afternoon. The asymmetry is not subtle.

The three RIAs in the orders did not get fined because they used AI badly. They got fined because the gap between the marketing copy and the evidence file was wide enough to fit a Wells notice through. Close the gap before the copy goes live and the entire enforcement risk profile collapses.

TRY THE TOOL
Run the free Marketing Rule preflight check at /ria/marketing-rule-preflight — upload one piece of AI marketing copy, get a per-claim substantiation gap report in 90 seconds.
Cite this article

Ascero AI. “SEC Marketing Rule + AI-Washing: 3 Mistakes That Got RIAs Fined.” May 28, 2026. https://asceroai.com/news/sec-marketing-rule-ai-washing-3-mistakes

Free to reference with attribution and a link back to this page.

Did this land? Pass it on.
Share X LinkedIn Email