The SEC and AI Disclosures: Potential Enforcement on the Horizon
November 16, 2026
On October 30, 2026, SEC Enforcement Director David Woodcock confirmed that the Division of Enforcement is actively scrutinizing public companies’ disclosures about artificial intelligence.[1]
His comments follow a year of signals from Chairman Paul Atkins that AI disclosures will be judged under traditional materiality principles, and they suggest the SEC’s focus is widening from overstated capabilities to whether companies adequately disclose material AI use, spending, and risk.
The SEC has not brought an AI disclosure action against a public company since early 2025, but courts have now issued the first wave of pleading-stage rulings in private securities cases related to AI disclosures, and those rulings are a useful guide to where the SEC is likely to look. The theories that have survived motions to dismiss illustrate potential applications of Director Woodcock’s stated disclosure focus, and the SEC, which is not bound by all of the pleading standards that defeat many private claims and can subpoena the internal records private plaintiffs lack, is well positioned to pursue those theories. This memo maps the private case law onto likely enforcement priorities and offers practical steps for the 2026 year-end disclosure cycle.
I. The SEC’s Signals
In interviews with Law360 and Reuters, Director Woodcock said the Division of Enforcement will scrutinize public company statements that exaggerate AI capabilities or that fail to disclose AI use or risks in a fundamental part of the business, following the template of earlier cases against private companies. “Our principal concern,” he said, “is that people are making accurate disclosures about material information so that investors know what they’re getting into.”[2] Notably, in addition to overstatement of AI capabilities (i.e., “AI-washing”), the Director’s remarks suggest scrutiny of two other categories of disclosure failures. The first is outright omission: a company that does not disclose a material dependence on AI, or material AI capital commitments, at all. The second is understatement: a company that discloses its AI use or spending but characterizes it in a way that minimizes its significance, so that the disclosure is misleading in context. Either can create exposure under the applicable disclosure and antifraud provisions.[3] These remarks suggest that scrutiny could extend beyond AI-washing to AI spending, financing structures, counterparty concentration, and the assumptions behind projected returns.
Chairman Atkins has consistently framed the SEC’s approach to public company disclosure in principles-based terms that emphasize financial materiality, rejecting a prescriptive, rules-based approach.[4] In line with that theme, Chairman Atkins has said that a principles-based approach rooted in materiality “should inform how a public company today ought to disclose developments concerning AI, just as it guides disclosures about any other development,” and that disclosure “checklists” are no substitute.[5] As a separate note on practice, Chairman Atkins has also cautioned that AI “should serve as a complement to—not a substitute for—human judgment,” a warning that applies to companies using generative AI to draft their own filings.[6] For firms within its examination program, the Division of Examinations’ fiscal year 2027 priorities likewise address emerging financial technologies, including whether representations are accurate and operations and controls are consistent with disclosures.[7]
The SEC’s January 2025 settled order against Presto Automation Inc. illustrates public-company AI disclosure risk. Presto, a restaurant technology company, had deployed an AI voice-ordering product, “Presto Voice.” The SEC found that Presto’s filings were misleading in that it described the AI speech-recognition technology in its Presto Voice units as its own without disclosing that, for a period of time, the technology in all units was owned and operated by a third party. The SEC also found Presto misleadingly disclosed the rate of orders completed without human intervention.[8] Notably, the SEC charged violations of the Securities Act Section 17(a)(2) antifraud provision and Exchange Act Section 13(a), including the disclosure-controls rule, none of which requires proof of scienter, that is, an intent to deceive. Other AI-washing matters have involved investment advisers[9] and private companies, including Nate, Inc., whose founder was the subject of parallel SEC and criminal charges[10] alleging that its purportedly AI-driven shopping app relied on human workers.
II. Private Litigation as a Roadmap for SEC Scrutiny
AI-related securities class actions have more than doubled since 2023. Several have been dismissed, but the rulings illustrate an important distinction: courts treat generic claims of AI “leadership” as puffery and reject AI-washing theories premised on the contention that a company had no real AI at all where the plaintiffs’ allegations show otherwise, while sustaining claims built on specific, verifiable statements about what a company’s AI does or what it is driving. Theories that were rejected for failure to plead particularized facts or scienter may fare differently in an SEC investigation, where the staff can obtain internal documents by subpoena and, under Sections 17(a)(2) and 17(a)(3) of the Securities Act and Section 13(a) of the Exchange Act, need not prove scienter.
Specific Claims About Model Capability and Performance
Courts have sustained claims premised on concrete descriptions of how an AI model works. In In re Upstart Holdings, Inc. Securities Litigation (S.D. Ohio), the court allowed claims to proceed based on specific statements that Upstart’s lending model delivered measurable advantages over traditional models and could “quickly adjust” to changing macroeconomic conditions, while rejecting as puffery descriptions of the model as a “fairly magical thing.”[11] In September 2026, the court allowed related securities fraud and insider trading claims to proceed against an early investor and its board designee.[12] In In re GigaCloud Technology Inc. Securities Litigation (S.D.N.Y.), the court allowed claims under Sections 11 and 15 of the Securities Act to proceed based on allegations that offering materials misdescribed the company’s use of AI in its logistics operations, citing alleged statements from former employees who described a largely manual system.[13] These are the same kinds of statements the SEC charged in Presto: quantifiable descriptions of what the technology does and how much human involvement it requires.
AI-Driven Demand, Revenue, and Customer Concentration
Claims that tie financial results to AI are emerging as a second front, and they illustrate how a general focus on AI disclosures could extend to AI-attributed revenue and customer concentration. In Lucid Alternative Fund, LP v. Five9, Inc. (N.D. Cal.), the court rejected the argument that statements that Five9 was “winning these large enterprise deals” because of its AI products were mere puffery, describing them as “potentially actionable,” but found that the plaintiff had not pleaded facts showing they were false (although the case has proceeded on other statements concerning backlog, sales, and macroeconomic effects).[14] The takeaway is that statements attributing growth or customer wins to AI are factual claims that can be tested, not protected opinion, and a plaintiff, or the SEC, that can show the attribution was unsupported may well prevail. Companies that make such attributions should expect the SEC to ask what supports them and whether known risks to the underlying relationships were disclosed.
SEC action involving earlier emerging technologies may prove especially informative. In the wave of electric vehicle cases it brought between 2021 and 2024, the SEC first targeted claims about the technology itself, such as Nikola’s statements about its technical advancements and in-house production capabilities[15] and Hyzon Motors’ claim that it had delivered a hydrogen fuel cell vehicle that was in fact not equipped to run on hydrogen.[16] Later cases turned to more nuanced misstatements about commercial prospects, including Lordstown Motors’ characterization of nonbinding preorders as evidence of fleet-customer demand[17] and Canoo’s allegedly unreasonable revenue projections.[18] A similar dynamic played out with cryptocurrency: the SEC started by targeting a variety of alleged crypto-related frauds, and then in 2022, the SEC charged Nvidia with failing to disclose in its Forms 10-Q that cryptomining was a significant factor in the growth of its gaming revenue, even as it attributed growth in other segments to crypto demand. Director Woodcock’s focus on AI spending and counterparty concentration suggests AI enforcement is now following the same arc, from whether the technology exists to whether its significance is accurately described.
Product Timelines and Undelivered Features
City of Coral Springs Police Officers Pension Plan v. Apple Inc. (N.D. Cal.) alleges that Apple promoted an iPhone upgrade “supercycle” built around Apple Intelligence and a “more personal” Siri it was not close to delivering. The case is stayed without a merits ruling,[19] but a consumer class action over the same features, Landsheft v. Apple Inc., produced a $250 million settlement that was preliminarily approved in July 2026.[20] By contrast, the Ninth Circuit affirmed dismissal of claims over Tesla’s autonomous-driving timelines, holding that Tesla’s repeated, specific warnings about development risk undermined any inference of scienter.[21] Any SEC inquiry related to timelines would likely turn on what internal readiness assessments showed at the time.
Where Courts Have Drawn the Line
The cited decisions illustrate two recurring limits on private claims. First, generic AI superlatives are puffery: Five9’s “leadership in AI” and GitLab’s stated belief that it was positioned as “the leading AI-powered DevSecOps platform” fell into this category.[22] Second, courts have rejected AI-washing claims premised on the theory that a company had no real AI at all where the plaintiffs’ own allegations showed some functioning AI capability. In D’Agostino v. Innodata Inc. (D.N.J.), the court dismissed for insufficient particularized allegations of falsity and scienter, noting that the alleged employee accounts described functioning AI. The opinion also notes that the SEC and DOJ had closed investigations without enforcement action.[23] In Sarria v. TELUS International (CDA) Inc. (S.D.N.Y.), the court rejected the theory that TELUS lacked AI offerings where the allegations acknowledged data-annotation services and pilot programs and failed to identify “any specific AI product or service” falsely claimed to exist.[24] These rulings do not mean that companies with any genuine AI product are insulated: specific misstatements about what that product does, how it performs, or how much human involvement it requires may support actionable claims, as Upstart, GigaCloud, and Presto show. These private-action dismissals do not necessarily foreclose SEC enforcement: Innodata, GitLab, and Tesla each turned in part on the absence of particularized facts or scienter, gaps an SEC investigation may be able to fill.
III. Takeaways for Market Participants
Substantiate specific claims. Statements about model performance, accuracy, automation rates, or the share of work done without human involvement are among the types of specific, measurable AI-related claims that have survived legal challenges and formed the basis of SEC enforcement actions. Such claims should be substantiated by contemporaneous, reliable data and accurately reflect the technology’s capabilities and actual deployment. Disclosures should distinguish proprietary models from third-party models, conventional automation from machine learning, and pilot programs or limited testing from scaled deployments.
Disclose AI economics, not just AI risks. Companies should put in place disclosure controls and procedures around AI disclosures to ensure that they consider whether AI capital expenditures, financing arrangements, customer or supplier concentration, AI-attributed revenue, or other AI-related metrics warrant disclosure, including whether they constitute known trends, risks, or uncertainties that should be disclosed in the Management’s Discussion and Analysis section of periodic reports. Companies that attribute growth to AI demand should consider whether known problems in the underlying relationships must be disclosed. Risk factors should be tailored and should not misleadingly present a materialized risk as merely hypothetical. Roadmaps should accurately describe current readiness and carry specific, current cautionary language; Tesla illustrates that such warnings can inform the scienter analysis, not that they excuse false statements about existing facts.
Disclosure controls should address AI-related disclosures. Presto included a disclosure-controls charge. Existing sub-certification and disclosure committee processes should include review of any claims about AI capabilities and any AI-related metrics, and companies should confirm that people with technical knowledge of the company’s AI products and use, such as engineering, product, or data science leads, are reviewing AI-related disclosures. Because AI claims often originate outside SEC filings, companies should also coordinate AI-related statements across filings, earnings scripts, investor presentations, and marketing materials, so that claims made in each channel are consistent and are vetted. Boards and/or their constituent committees with relevant oversight responsibility, such as audit committees, should address material AI-related disclosure matters as part of their regular oversight of the company’s disclosure processes, with access to relevant technical expertise as appropriate. Finally, where generative AI tools are used to help draft filings, their output should be subject to human verification.
Watch for red flags. Certain claims related to AI are more likely to prompt regulatory and investor scrutiny than others. Examples include claims about a pivot to AI without a corresponding change in operations, revenue, or technology. Quantified claims about AI performance or automation that are not supported by data are another example. The same is true of promoting AI features as available or ready for deployment before they actually are. Each of these may raise particular scrutiny and increase litigation or enforcement risk. Companies should consider these risks in crafting their public disclosures.
Cleary Gottlieb’s Artificial Intelligence Enforcement and Investigations practice,[25] together with our capital markets, public company advisory, and securities litigation colleagues, has broad experience advising clients across a range of emerging enforcement risks related to artificial intelligence, including new federal and state investigative efforts and related compliance assessments. We are available to assist with AI disclosure reviews, governance assessments, and the response to any SEC inquiry or shareholder suit.
* Cleary associate Adrian Gariboldi also contributed to this alert.
[1] Chris Prentice, US SEC’s Top Cop Says Agency Is Monitoring AI Corporate Disclosures, Reuters (Oct. 30, 2026), https://www.reuters.com/legal/government/us-secs-top-cop-says-agency-is-monitoring-ai-corporate-disclosures-2026-10-30/; Law360, The SEC Is Reading Your Corporate AI Disclosures (Oct. 30, 2026), https://www.law360.com/articles/2544508.
[2] Chris Prentice, US SEC’s Top Cop Says Agency Is Monitoring AI Corporate Disclosures, Reuters (Oct. 30, 2026), https://www.reuters.com/legal/government/us-secs-top-cop-says-agency-is-monitoring-ai-corporate-disclosures-2026-10-30/.
[3] See Macquarie Infrastructure Corp. v. Moab Partners, L.P., 601 U.S. 257 (2024) (pure omissions are not actionable under Rule 10b-5(b), but the SEC retains authority to enforce its disclosure rules, including Item 303 of Regulation S-K).
[4] See, e.g., Chairman Paul S. Atkins, Remarks at the Society for Corporate Governance Conference (July 9, 2026), https://www.sec.gov/newsroom/speeches-statements/atkins-remarks-society-corporate-governance-07-09-2026-remarks-society-corporate-governance-conference; Chairman Paul S. Atkins, Remarks at the Texas A&M School of Law Corporate Law Symposium (Feb. 17, 2026), https://www.sec.gov/newsroom/speeches-statements/atkins-02-17-2026-remarks-texas-am-school-law-corporate-law-symposium.
[5] Chairman Paul S. Atkins, Remarks at FSOC Artificial Intelligence Innovation Series Roundtable on Strategy and Governance Principles (Mar. 4, 2026), https://www.sec.gov/newsroom/speeches-statements/atkins-remarks-at-financial-stability-oversight-council-artificial-intelligence-innovation-series-roundtable-030426.
[6] Chairman Paul S. Atkins, Remarks at the Investor Advisory Committee Meeting (Sept. 10, 2026), https://www.sec.gov/newsroom/speeches-statements/atkins-remarks-iac-091026.
[7] SEC Division of Examinations, Fiscal Year 2027 Examination Priorities at 12 (Oct. 23, 2026), https://www.sec.gov/files/2027-exam-priorities.pdf; SEC Press Release 2026-109 (Oct. 23, 2026), https://www.sec.gov/newsroom/press-releases/2026-109-sec-announces-fiscal-year-2027-examination-priorities.
[8] SEC, SEC Charges Restaurant-Technology Company Presto Automation for Misleading Statements About AI Product (Jan. 14, 2025), https://www.sec.gov/enforcement-litigation/administrative-proceedings/33-11352-s.
[9] SEC Press Release 2024-36 (Mar. 18, 2024), https://www.sec.gov/newsroom/press-releases/2024-36.
[10] SEC Litigation Release No. 26282 (Apr. 11, 2025), https://www.sec.gov/enforcement-litigation/litigation-releases/lr-26282; U.S. Attorney’s Office, S.D.N.Y., Tech CEO Charged in Artificial Intelligence Investment Fraud Scheme (Apr. 9, 2025), https://www.justice.gov/usao-sdny/pr/tech-ceo-charged-artificial-intelligence-investment-fraud-scheme.
[11] In re Upstart Holdings, Inc. Sec. Litig., No. 2:22-cv-02935, 2023 WL 6379810 (S.D. Ohio Sept. 29, 2023), ECF No. 68, at 23–29.
[12] In re Upstart Holdings, Inc. Sec. Litig., No. 2:22-cv-02935, Opinion & Order at 25–38, 45–61, ECF No. 181 (S.D. Ohio Sept. 30, 2026).
[13] In re GigaCloud Tech. Inc. Sec. Litig., No. 1:23-cv-10645, 2025 WL 307378 (S.D.N.Y. Jan. 27, 2025).
[14] Lucid Alternative Fund, LP v. Five9, Inc., No. 24-cv-08725-PCP, Order Denying Motion to Dismiss at 15–17, 24, ECF No. 58 (N.D. Cal. Feb. 23, 2026).
[15] SEC Press Release 2021-267, Nikola Corporation to Pay $125 Million to Resolve Fraud Charges (Dec. 21, 2021), https://www.sec.gov/newsroom/press-releases/2021-267.
[16] SEC Press Release 2023-200, SEC Charges Hydrogen Vehicle Co. Hyzon Motors and Two Former Executives for Misleading Investors (Sept. 26, 2023), https://www.sec.gov/newsroom/press-releases/2023-200.
[17] SEC Press Release 2024-29, SEC Charges Lordstown Motors with Misleading Investors about Company’s Flagship Electric Vehicle (Feb. 29, 2024), https://www.sec.gov/newsroom/press-releases/2024-29.
[18] SEC Litigation Release No. 25802, SEC v. Kranz, No. 23-cv-06332 (C.D. Cal. Aug. 4, 2023) (civil charges against former executives and a separate settled administrative proceeding against Canoo Inc.), https://www.sec.gov/enforcement-litigation/litigation-releases/lr-25802.
[19] City of Coral Springs Police Officers Pension Plan v. Apple Inc., No. 25-cv-06252-NW, Order Staying Action at 2, 4–5, ECF No. 114 (N.D. Cal. July 17, 2026); Consolidated Amended Complaint ¶ 3, ECF No. 65 (Jan. 28, 2026).
[20] Landsheft v. Apple Inc., No. 5:25-cv-02668-NW, Order Granting Preliminary Approval, ECF No. 94 (N.D. Cal. July 17, 2026); Settlement Agreement § III.B.1, ECF No. 78-1 (filed May 5, 2026) ($250 million total settlement amount).
[21] Oakland Cnty. Voluntary Emps.’ Beneficiary Ass’n v. Tesla Inc., No. 25-55, 2025 WL 3459471 (9th Cir. Dec. 2, 2025), memorandum at 2–6 (unpublished; nonprecedential except as provided by Ninth Circuit Rule 36-3).
[22] Dolly v. GitLab Inc., No. 24-cv-06244-EKL, Order Granting Motion to Dismiss at 15, ECF No. 72 (N.D. Cal. Aug. 14, 2025); Final judgment (N.D. Cal. Jan. 26, 2026).
[23] D’Agostino v. Innodata Inc., No. 2:24-cv-00971, Opinion at 14, 23–27, 35–36, ECF No. 58 (D.N.J. Apr. 27, 2026).
[24] Sarria v. TELUS Int’l (CDA) Inc., No. 1:25-cv-00889, Decision and Order at 41–46, ECF No. 43 (S.D.N.Y. Aug. 11, 2026).
[25] Cleary Gottlieb, Artificial Intelligence Enforcement and Investigations.