The biggest firms disclose AI use four times as often as the smallest
Disclosed AI use appears in 23.7 percent of Form ADV Part 2A brochures filed by investment advisers registered with the US Securities and Exchange Commission, while 27.6 percent of the same documents carry AI-related risk language.
Independent New York City researchers Samir Chincholikar and Robin Chawla put the within-brochure difference at 3.9 percentage points with an exact McNemar p value of 0.036 in their study of 388 brochures, published in October in Frontiers in Artificial Intelligence.
Of the 388 brochures, 77 carry both risk language and disclosed use, 30 carry risk language only, 15 carry disclosed use only, and 266 carry neither.
Against a measure that excludes named-vendor references, the gap widens to 5.2 percentage points at p = 0.007.
Disclosure rises across the 2024 assets under management quartiles, from 10.4 percent in the smallest to 43.8 percent in the largest.
Private-fund advisers disclosed use in 31.5 percent of brochures against 15.4 percent among wealth and retail advisers.
In a logistic regression, the coefficient on AUM quartile is 0.67 and the coefficient on private-fund status is 1.01, both at p < 0.001, which the authors describe as descriptive associations rather than causal estimates.
The same corpus yields estimates of 9.9 percent, 13.1 percent, 23.7 percent and 31.7 percent depending on the construct, adviser universe and weighting applied.
Restricting the measure to brochures containing the literal phrase “artificial intelligence” and weighting to the wealth and retail population gives 9.9 percent; the broadest any-mention measure gives 31.7 percent.
A literal-phrase keyword rule achieves precision of 0.75 and recall of 0.74 against the study’s classifier on identical text.
AI-disclosure prevalence “reflects measurement design as much as adviser behavior,” Chincholikar and Chawla write in Frontiers in Artificial Intelligence.
The sample was drawn from 16,223 active SEC-registered advisers reporting regulatory AUM as of 31 December 2024, with 50 firms selected from each of eight strata crossing adviser type and AUM quartile.
GPT-4o classified the brochures under a five-label typology.
A blinded recoding of 180 brochures by Anthropic’s Claude returned precision of 0.76 and a Cohen’s kappa of 0.74 on the composite any-use measure.
Against 120 human-coded brochure cases, the primary classifier returned precision of 0.76, sensitivity of 0.81 and specificity of 0.94.
A screen built from the SEC’s two March 2024 AI-washing enforcement orders flagged five of the 388 brochures, or 1.3 percent.
None was adjudicated as an unsubstantiated promotional claim.
Those orders concerned Delphia (USA) Inc. and Global Predictions Inc., which agreed to civil penalties of US$225,000 and US$175,000 respectively, the paper states.
Across 283 firms observed in both venues, 23.0 percent of brochures and 5.7 percent of principal websites contained disclosed AI use.
The matched brochures carry 7.3 times the characters of the matched websites, and the 17.3-percentage-point venue difference is no longer distinguishable from zero once log document length enters the model.
The Charles Schwab 2026 RIA and AI research study found that about 63 percent of surveyed firms are using or piloting AI, a figure the authors treat as a benchmark divergence rather than a nondisclosure rate, because the two instruments measure different constructs on different populations at different dates.
The study covers US SEC-registered advisers only.
In Canada, the Canadian Securities Administrators published Staff Notice and Consultation 11-348 on December 2024, stating that registrants must disclose the use of AI systems in ways that may directly affect registerable services provided to clients, in registration applications or change filings.
The notice says conduct involving inaccurate, false, misleading or embellished AI claims “may be misleading to the public or constitute a misrepresentation,” as defined by securities legislation.
On discretionary portfolio management, CSA staff wrote that it would be challenging for a registrant using a fully autonomous AI system to demonstrate proper compliance with securities laws, and that at the current stage of development they do not believe an AI system can substitute for an advising representative acting as decision-maker.
