The SEC Now Checks If Your AI Claims Are Actually True

In 2023, 12% of S&P 500 companies listed AI as a material risk in their annual filings. Today that number is 72%. That jump happened fast, under pressure to look current. Now someone is checking the answers.

The SEC's Division of Examinations listed AI as a priority in its 2026 exam guidance. It said it will "review for accuracy registrant representations regarding their AI capabilities." That is a different commitment than checking whether disclosure exists. Examiners are now asking whether what you wrote is actually true.

What changed about how the SEC looks at AI disclosures in 2026?

The SEC is not writing new AI-specific rules. It is applying the same principles-based standards it has always used — and pointing them directly at AI claims.

No new regulation is needed for this. SEC Chairman Paul Atkins has stated that existing rules already allow companies to report on material AI impacts. The old rules are sufficient. What changed is how seriously the agency is checking the answers.

Law firm Goodwin Law's December 2025 analysis of the 2026 priorities concluded that AI oversight will be "a component of virtually all examinations going forward." That language covers investment advisers, broker-dealers and public registrants. It is not limited to tech companies or large firms.

You used to have to say something about AI. Now you have to say something accurate. That is the change.

What is AI washing — and why is the SEC treating it as a problem?

AI washing means overstating your AI tools' capabilities, autonomy or proprietary nature in public filings, Form ADV brochures or investor marketing materials.

The name follows the same logic as greenwashing. You do not have to lie outright. Overstating is enough. Then-SEC Chair Gary Gensler put the standard plainly: companies must "say what they're doing and do what they're saying." That applies whether the overstatement was intentional or just marketing language no one fact-checked against the actual system.

Three real enforcement cases your clients should know about

The SEC has already charged and settled with multiple companies. These are not hypothetical risks.

Each case follows the same pattern. The AI existed. The company described it as doing more than it did. Examiners found the gap between description and reality. That gap was the violation.

Does vague language protect your clients?

Vague language does not protect a company. Generic phrases like "we leverage AI for efficiency" are exactly what examiners are trained to identify and flag.

Language like "we leverage AI to improve operational efficiency" does not protect a company. It can actually create a problem. The SEC's position, summarized in Paul Hastings' January 2026 client alert, is that companies should avoid language suggesting AI is "more advanced, capable, autonomous or proprietary than it is."

Vague language often overclaims by implication. If a filing says AI is driving outcomes and a human is actually doing the work, that statement may still be misleading — even if the word "AI" is never defined or quantified. The safer path is specific language that names the tool, describes what it does and says what it does not do.

That takes more work during the filing review. It also holds up when examiners ask.

Which clients and filing types face the most scrutiny?

Investment advisers and public companies face the highest immediate risk. The 2026 exam priorities address them directly.

Investment advisers and Form ADV

The SEC will examine whether investment advisers' actual AI usage matches what their Form ADV Part 2A says. Examiners will assess written policies under Rule 206(4)-7 — whether those policies address AI accuracy, confidentiality, recordkeeping and bias, and whether they are operating in practice rather than just written down.

If a client's brochure says AI analyzes market signals for portfolio decisions, that system needs to be running. If it was replaced by a different process after the last filing, the Form ADV needs to reflect that.

Public companies and Form 10-K material risk factors

The numbers tell you how much exposure there is. Research by The Conference Board and ESGAUGE found that 72% of S&P 500 companies now disclose AI as a material risk in their public filings, up from 12% in 2023.

That jump means a large number of companies added AI language to their filings quickly, under pressure to show they were current. Some of those statements will not survive a factual review. Reputational risk from AI failures is the most frequently cited concern — 38% of companies flagged it in 2025 filings, according to The Conference Board study.

What should your firm add to its pre-filing review process?

The filing cycle is the right time to catch accuracy problems before they become enforcement problems.

Before a client submits a 10-K or amends a Form ADV, add an AI claims review step to the checklist.

Four steps that hold up under examiner scrutiny:

1. Pull last year's filing. Find every sentence that uses "AI," "artificial intelligence," "machine learning" or "automation" in a capabilities or risk context.
2. Confirm documentation exists. For each claim, ask the client to produce evidence the system is actively deployed. Not a roadmap, not a demo. A running system with usage records.
3. Flag capability language. Phrases like "our AI decides," "our AI monitors in real time" or "AI-driven" require specific technical support. If that support does not exist, the language needs to change before filing.
4. Replace boilerplate with specifics. "We use AI to improve efficiency" needs to name what it improves, in what process and by what measure. The SEC wants language a reasonable investor can evaluate — not marketing copy carried over from a press release.

None of this requires a technical AI audit. It requires asking clients to show their work. That is already part of what a careful advisor does at filing time. The SEC just made it more consequential.

Before filing season, check two things. First, does your firm have a written AI policy that covers which tools are approved, what client data can enter them, and who reviews the output? If not, that gap is harder to explain during an examination than a missing statistic. See what that policy should cover for small accounting firms. Second, if you are still building your firm's AI adoption process, the implementation roadmap covers the governance controls that belong in place before you scale any AI workflow.