AI Washing Is the New Greenwashing, and the Securities Bar Is Already Filing

51 AI-related securities class actions have been filed in the last five years, and a significant majority allege that companies overstated or misrepresented their artificial intelligence capabilities. That data, compiled by the consulting firm Secretariat and published this week by Fortune, is the clearest numerical picture yet of where AI litigation risk actually lives. It is not in the product liability lane. It is in the 10-K.

The pattern now has a name: AI washing. The SEC already brought its first two enforcement actions against Delphia (USA) Inc. and Global Predictions Inc. in March 2024, with allegations that both firms marketed AI capabilities they could not substantiate. One of them claimed to be “the first regulated AI financial advisor.” The Commission found that claim unsupported. Both firms settled.

Two years later, the plaintiffs’ bar has caught up, and the volume is the story.

The Innodata case is the template

The data engineering firm Innodata, Inc. offers the clearest example of how this plays out. In February 2024, a short seller accused Innodata of exaggerating the role of AI in its business. Hagens Berman filed a securities class action. Innodata’s stock dropped 30 percent. The complaint alleges violations of Sections 10(b) and 20(a) of the Securities Exchange Act, the standard 1934 Act fraud claims, built on disclosures that Innodata’s “proprietary, state-of-the-art” Goldengate AI platform was materially less sophisticated than the company represented.

Note what the case does not require. The plaintiffs do not need to prove Innodata had no AI. They only need to show the disclosures overstated what the AI actually did, or misrepresented the degree to which AI drove the company’s revenue and margins. That is a much easier case to plead, and a much harder one to get dismissed at the 12(b)(6) stage.

Evolv Technologies is facing a similar case over its AI-powered weapons detection systems. More will follow.

The shift from “does the AI exist” to “does it matter”

Early AI washing cases looked like traditional fraud. Short sellers and plaintiffs’ firms argued that the defendant’s AI was vaporware, that the product marketed as machine learning was actually a rules-based script. Those cases still exist, but they are no longer the majority.

The more dangerous cases now ask a different question. The AI exists. The models are real. Does the AI meaningfully change the economics of the business? Does it improve margins, increase revenue, or create a defensible moat? If the company’s 10-K says yes and the operational reality says no, that is actionable under the federal securities laws.

This is the “ESG playbook” applied to AI. Greenwashing started as companies claiming to be sustainable when they were not. It matured into companies whose sustainability was real but immaterial to the financial thesis. The SEC and private plaintiffs now litigate the materiality gap on ESG. They are doing the exact same thing on AI.

Why GCs should be reading this before the next earnings cycle

A public company CFO and CEO sign certifications under Sarbanes-Oxley Section 302 attesting that the 10-K does not contain material misstatements. Every earnings call contains forward-looking statements about the company’s AI strategy. Every investor deck now has a slide called something like “Our AI Moat.”

If any of those statements overstate the operational reality, a plaintiff’s firm can plead scienter and fraud on the market. And they will. These cases now have a developed bar, templated complaints, and a five-year track record of surviving motions to dismiss.

The specific language that creates exposure:

• “Proprietary AI” when the underlying model is a fine-tuned open-source model that any competitor can replicate.
• “AI-driven” revenue when the AI is a downstream analytics layer on top of revenue that would have been generated regardless.
• “First” or “only” claims about AI capabilities when competitors have equivalent functionality.
• Machine learning claims where the actual system is rules-based automation.
• Customer adoption or usage metrics that conflate AI features with general product usage.

Any of these, stated in a 10-K, 10-Q, S-1, proxy, or earnings call, can be pulled into a securities complaint.

The private equity angle nobody is talking about

Fortune noted something buried halfway down the piece that deserves its own article. PE firms are currently deploying capital into AI-adjacent targets at ambitious valuations. The due diligence environment is compressed. The pressure to win deals is intense. Technical AI diligence requires specialized expertise that traditional deal teams do not always have.

PE firms that pay premium multiples for AI capabilities that turn out to be experimental, limited, or economically immaterial are going to be the plaintiffs in the next wave of these cases. Not the defendants. They are going to be suing the founders, the selling management teams, and the sell-side bankers under reps and warranties claims and securities fraud theories.

If you run M&A for a mid-market PE firm and your targets are pitching AI capabilities, your diligence process needs a dedicated AI technical reviewer. Not a generalist. Someone who can evaluate model architecture, training data quality, and deployment infrastructure.

What to do now

For public company GCs:

• Audit every forward-looking AI statement in your last four quarterly filings and two most recent investor presentations. Compare each statement to what your CTO can actually demonstrate in production. Any gap is a litigation target.
• Add AI-specific disclosure review to your quarterly SOX process. Your disclosure committee should have a standing agenda item on AI claims.
• If you have announced AI capabilities that are “coming soon” or “in development,” confirm that your disclosures make clear they are prospective. Safe harbor language matters, but it only works if the language is actually there.

For private company counsel preparing for an S-1 or SPAC transaction:

• Assume every AI claim in your registration statement will be litigated within 18 months of the IPO. Draft accordingly.
• Your product marketing team and your SEC counsel need to agree on a single vocabulary for AI capabilities. The homepage cannot say one thing and the S-1 another.

For PE and VC diligence teams:

• Add a technical AI diligence workstream to every deal where AI is part of the value thesis. This is not optional anymore.
• Include specific reps from management on what the AI does and does not do, with materiality thresholds. You want the ability to claim breach of reps later if the AI turns out to be immaterial.

The dot-com era ended with Sarbanes-Oxley. The ESG era produced the Names Rule and enhanced climate disclosures. The AI era is going to produce its own regulatory response, and the plaintiffs’ bar is not waiting for it. They are already filing, already surviving motions to dismiss, and already collecting settlements.

Every company that is going to need a securities lawyer in the next 24 months is a company whose AI claims can be pressure-tested against operational reality right now. Do the pressure test yourself before a Hagens Berman short-seller report does it for you.