Why Delphia and Global Predictions Lost $400,000 to AI Washing — And What Their Logs Should Have Shown
SEC Division of Enforcement
The Headline
On March 18, 2024, the SEC charged two investment advisors — Delphia (USA) and Global Predictions Inc. — with making false and misleading statements about their use of artificial intelligence. Delphia paid $225,000. Global Predictions paid $175,000. Combined: $400,000 in civil penalties.
That is the public story. What the SEC orders do not fully explain is how easily this could have been prevented — not with a different marketing team, but with a different architecture.
What the SEC Actually Alleged
Delphia claimed in SEC filings, press releases, and its website that it "put[s] artificial intelligence to work" to inform its investment strategies. The SEC found that Delphia did not have the AI capabilities it described. Its AI use was minimal and did not support the claims.
Global Predictions claimed to be the "first regulated AI financial advisor" and touted an "expert AI-driven forecast" that could predict market events up to three years in advance. The SEC found that Global Predictions had no reasonable basis for these claims and that its AI was not capable of the forecasting described.
Both firms violated Section 206(4) of the Investment Advisers Act and Rule 206(4)-7, which prohibit misleading statements and require compliance policies and procedures reasonably designed to prevent such violations.
Source: SEC Press Release 2024-25; Delphia SEC Order; Global Predictions SEC Order
The Compliance Failure, Not the Marketing Failure
The $400,000 penalty was not for bad AI. It was for unprovable claims.
The SEC did not ask "Was the AI good?" It asked: "What did you claim? What did the AI actually do? Can you show us the difference?"
Neither firm could. Not because they were hiding something, but because their infrastructure could not produce the evidence. They had marketing claims. They had AI systems. They had no immutable bridge between the two.
This is the gap that kills AI governance programs: the moment when a regulator asks for proof, and the firm produces a dashboard screenshot.
What the Logs Should Have Shown
If either firm had deployed forensic-grade inference logging — what RTFCT calls the Open Evidence Standard (OES) — the SEC's examination would have looked different.
An OES-compliant log for an AI-driven investment recommendation includes:
- Input provenance: What data fed the model? Market data, analyst reports, economic indicators? Where did it come from? When was it retrieved?
- Model identity: Which model version generated the recommendation? What were its training parameters? Was it fine-tuned on proprietary data, or was it a general-purpose model?
- Inference chain: What prompt produced the output? What was the raw model response? What post-processing was applied?
- Human review: Who reviewed the AI output? What changes were made? What was the rationale for accepting or rejecting the recommendation?
- Client delivery: What was ultimately communicated to the client? How did it differ from the raw AI output?
- Cryptographic integrity: Is the log tamper-evident? Can the firm prove the log was not altered after the fact?
Under FRE 902(13) and (14), a properly maintained OES log is self-authenticating. The firm does not need a witness to testify to its accuracy. The cryptographic proof speaks for itself.
The Lesson for GCs and CISOs
The Delphia and Global Predictions orders are not outliers. They are the opening act.
The SEC's 2024 AI washing cases are the first in what will be a long enforcement arc. The SEC has proposed rules for predictive data analytics (July 2023 NPRM) that would require broker-dealers and investment advisors to address conflicts of interest in AI-driven interactions. The CFPB has issued circulars on AI in adverse action notices. State regulators from Colorado to Texas are building AI-specific enforcement frameworks.
The question is not whether your firm will be examined. It is whether your infrastructure can answer the examiner's questions before they ask them.
The RTFCT Diagnostic
Every Statutory Autopsy ends with a next step. This one is simple:
Run the AI Disclosure Gap Diagnostic on your current AI deployment. We evaluate your marketing claims, your model capabilities, and your logging infrastructure against the SEC's AI washing enforcement framework. You receive a gap analysis — with specific remediation steps — in 72 hours.
Cost: Free for qualified in-house legal and compliance teams.
Deliverable: A confidential memo identifying which of your AI claims are provable, which are not, and what infrastructure is required to bridge the gap.