Securities Regulation Daily Wrap Up, FRAUD AND MANIPULATION—D.N.J.: AI-washing case against Innodata dismissed, (Apr 29, 2026)

By Mark S. Nelson, J.D.

The court said the complaint was sometimes correct, but its concessions about the company having some AI developments undercut any fraud allegations.

A proposed securities fraud class action brought against Innodata Inc. was dismissed via an unpublished opinion because the complaint, despite making some plausible allegations, ultimately was self-defeating because it conceded that the company, contrary to the gist of the overarching allegations of fraud, did have some type of AI product development and that concession took the complaint out of the fraud category. More specifically, the court found that the complaint failed to allege material misstatements or omissions and also failed to allege scienter or loss causation. The case was dismissed without prejudice (D’Agostino v. Innodata, Inc., No. 2:24-cv-00971-JKS-JSA (D.N.J. Apr. 27, 2026)).

According to the complaint, statements by Innodata executives touted the company’s ability to use artificial intelligence (AI) to enhance products offered to customers. The complaint sought to tell a story that Innodata claimed to have first internally validated its AI prowess and then turned that prowess into customer-facing applications. However, according to the complaint and a short-seller report cited in the complaint, Innodata instead propped up its AI program with labor-intensive human oversight that undermined public claims made to investors by company executives. But, as the court detailed, the complaint itself unraveled by making key concessions about Innodata’s AI program.

Materiality. Innodata asserted that the complaint should be dismissed because it indulged in forbidden puzzle pleading and relied on confidential witnesses (former Innodata employees) whose statements were unreliable for multiple reasons. Innodata also claimed that it had adequately disclosed limits regarding its AI capabilities and related financial capabilities.

The court rejected the puzzle pleading assertion because the complaint had sufficient structure to state the key components of materiality under the Private Securities Litigation Reform Act (PSLRA), including the who, what, when, and where of the statements cited.

The court also addressed Innodata’s assertion that the company had made adequate disclosures of key matters alleged in the complaint to have been deficient. The court noted that Innodata did not directly address the complaint’s “core” allegations about AI washing or unsubstantiated AI touting efforts by the company but instead cited disclosures about the ongoing role of humans and budgetary limits that prevented the company from competing with the largest AI-based companies.

With respect to the confidential witnesses cited in the complaint, the court said, in part, that the complaint lacked particularity because the cited witnesses did not remain employed with Innodata during the entire class period. A report titled the “Wolfpack Report,” published by former employees was, according to the court, deficient in some respects but otherwise sufficiently particular regarding allegations based on independent research.

“The former employees quoted in the report are not identified, their roles are not specified, the time periods of their tenure at Innodata are not specified, and the bases of their knowledge are thus not established in conformity with the ‘who, what, when, where, and how’ requirements of the PSLRA,” said the court.

Scienter. With respect to scienter, the complaint alleged that stock sales by Innodata executives were suspicious, that those same executives would have had knowledge of various AI programs and the results of such programs, and that executives sought to delay disclosure of SEC and DOJ investigations in order to soften bad news for investors.

Innodata responded that its executives held much of their company stock throughout the class period, that the complaint did not explain its theory about the timing of stock sales, and that the complaint cited stock sales by persons who were not made parties to the case. The company also asserted that the complaint was wrong about executives’ motives in disclosing the investigations when they did.

According to the court, the complaint failed to plead a strong inference of scienter. First, the court said the complaint failed to allege facts showing that company executives had the motive and opportunity to mislead investors. In the case of some company statements challenged by the complaint, the complaint itself conceded that Innodata had “some kind of AI” and, thus, company statements about AI must have had some degree of truth. Lastly, the court said the complaint failed to plead that the company had a duty to disclose the several investigations.

Moreover, the complaint separately did not plead scienter against a former part-time Innodata CFO. Said the court: “Spelker’s invocation of In re Advanta and similar cases is apt, as Defendant [sic] [plaintiff] pleads no other facts supporting a strong inference of scienter besides the bare assertion that Defendant Spelker’s position as part-time CFO would have apprised him of the alleged falsity of Innodata’s statements about AI. This is not enough to sustain a securities fraud claim.”

Loss causation. The loss causation section of the opinion focused on whether the Wolfpack Report was a corrective disclosure. Innodata said it was not because the report contained disclaimers about its accuracy and completeness and because the report merely offered opinions about previously disclosed facts rather than revealing new facts.

The plaintiffs, by contrast, argued that the report was a corrective disclosure because its publication was followed by a drop in Innodata’s share price. The plaintiffs also argued that Innodata was attempting to persuade the court to inappropriately apply non-circuit law regarding loss causation.

The court agreed with the plaintiffs that, in the Third Circuit—as opposed to the Ninth Circuit—corrective disclosures are judged under the lower short-and-plain-statement-of-the-case standard under FRCP 8(a) (i.e., “economic loss and its causal connection to the alleged misrepresentations and/or omissions”) rather than under the heightened pleading standards under FRCP 9(b) and the PSLRA. However, the court ultimately sided with Innodata because the complaint undermined itself by alleging that the company had some type of AI and, thus, failed to allege facts showing that Innodata engaged in misrepresentations or omissions that caused plaintiffs’ losses.

The case is No. 2:24-cv-00971-JKS-JSA.

Judge: Semper, J.

Attorneys: Donald A. Ecklund (Carella, Byrne, Cecchi, Olstein, Brody & Agnello, PC) for David D'Agostino. Harvey Bartle, IV (Morgan Lewis & Bockius LLP) for Innodata Inc.

Companies: Innodata, Inc.

LitigationEnforcement: AINews FedTracker Securities FraudManipulation GCNNews InvestorEducation PublicCompanyReportingDisclosure RiskManagement NewJerseyNews