INV-11 · AI-washing audit

Find "AI-powered" claims in public fund or company marketing and test each one against what the organization's own public filings and documents support. What would an honest disclosure standard require?

Public marketing pages and the same organizations' public filings, each captured with date accessed. Start with these primary sources, then look for others: SEC press release 2024-36 (https://www.sec.gov/newsroom/press-releases/2024-36), the two settled orders it announced (https://www.sec.gov/files/litigation/admin/2024/ia-6573.pdf) and (https://www.sec.gov/files/litigation/admin/2024/ia-6574.pdf), Chair Gensler's March 18, 2024 statement (https://www.sec.gov/newsroom/speeches-statements/sec-chair-gary-gensler-ai-washing), and the joint investor alert from the SEC's Office of Investor Education and Advocacy, NASAA, and FINRA (https://www.investor.gov/introduction-investing/general-resources/news-alerts/alerts-bulletins/investor-alerts/artificial-intelligence-fraud). Cite the regulator's own document, not a commentary about it. If a jurisdiction has no such document, the packet says so. These five are a starting set, not a complete list. AUDIT: any model used to summarize claims is itself audited.

Treating marketing language as evidence. Calling a claim false when it is only unsupported. Citing commentary about regulation instead of the regulator's own documents. Naming an organization as dishonest. Writing a standard nobody could follow.

A claims register (claim, location, date, supporting evidence or its absence, and a mark: supported, unsupported, or contradicted); the regulator sources found, cited at the primary source; and a one-page disclosure standard for "AI-powered" claims, with two examples rewritten to meet it.

CT: separating "unsupported" from "false." CR: drafting a standard that is short and testable. CO: a peer rates ten claims blind and the two registers are reconciled. CM: the disclosure standard. CZ: investors who buy on a claim that is not true.

A compliance professional or attorney reviews the standard. Standing instruction: reject any packet framed as legal advice or naming an organization as dishonest. School-supervised.

R1: one sector, one claim type. R2: every claim captured with date. R3: comparator is the organization's own filings. R4: limits of public evidence named. R5: the standard is usable. R6: no accusation; not legal advice.

(a) A list of companies called liars. (b) A standard so vague it permits every claim it was written to stop.

Lane A (government, economics, or computer science capstone). Pairs with Algorithmic Policy. No verified credit claim.