Two executives charged with fabricating the customers behind a $420 million AI revenue claim

What is actually alleged

Puthugramam Chidambaran, 57, founded iLearningEngines in Bethesda, Maryland in 2010. Sayyed Farhan Ali Naqvi, 44, joined as chief financial officer in 2019. The company sold itself as an out-of-the-box AI platform that let customers productise their institutional knowledge, and it reported revenue climbing from $217.9 million in 2021 to $420.6 million in 2023. On 17 April 2024 it completed a reverse merger with a special purpose acquisition company called Arrowroot and began trading on the Nasdaq under the ticker AILE. It reached a market capitalisation of roughly $1.5 billion and borrowed $60 million from the New York branches of two financial institutions.

Prosecutors say almost none of the revenue existed. The indictment alleges that between January 2019 and April 2025 the two men, working with five unnamed co-conspirators, inflated revenue by hundreds of millions of dollars a year, at times representing more than 90 percent of the annual figure. The mechanism was a set of sham contracts with entities that iLearning employees, their relatives, and their friends owned or controlled. Some agreements were signed by employees or employees’ family members posing as executives of the purported customer. In at least one instance a website was built for a shell entity so that an investor checking on it would find something. Money moved in a circle. Funds raised from investors and lenders went out to these customers and came back as licence payments, and the indictment puts the aggregate value of those round trips above $144 million. Chidambaran received more than $500 million in stock from the going-public transaction and about $12.5 million in restricted stock units. Naqvi received about $11.2 million in stock and nearly $4.5 million in cash to cover the tax on it. Both men are presumed innocent.

Where the artificial intelligence was, and was not

Nothing in the indictment alleges that artificial intelligence was used to commit any part of this offence. No model generated the fake executives. No voice clone called the auditor. The shell company website is described as a website. The forged customer relationships were built the way forged customer relationships have always been built, by finding people willing to sign things and lie about them afterwards.

What the AI claim did was supply the story that made the numbers plausible. A software company reporting 90 percent year-over-year growth invites questions. A company positioned as an early mover in enterprise AI in 2023 did not, or at least did not invite enough of them, because the sector’s economics were genuinely strange and rapid growth was genuinely happening elsewhere. The claim also explained the opacity. iLearning routed most of its business through an unnamed technology partner and described that arrangement as ordinary channel structure. In a sector that investors understood poorly, an unexplained intermediary in an offshore jurisdiction read as complexity rather than concealment. That is what the AI branding bought, and it is the only sense in which this case involves the technology at all.

The short seller found it first

On 29 August 2024 Hindenburg Research published a report titled iLearningEngines: An Artificial Intelligence SPAC With Artificial Partners And Artificial Revenue. It identified the unnamed technology partner as Experion Technologies and set out the overlap between the two companies, including that Chidambaran had at one point been listed as the American contact for Experion at his own home address. It estimated that iLearning’s Indian subsidiary had reported roughly $853,471 in revenue against a claimed regional run rate of $216 million. The stock fell 53.3 percent that day, from $3.19 to $1.49.

The company denied it, formed a special committee, and published a rebuttal. Then the denials unwound in stages. On 18 November 2024 iLearning told the market that its financial statements going back to 2020 should no longer be relied upon, that its auditor had withdrawn its opinions, and that it had received an SEC subpoena. On 10 December five executives were placed on administrative leave. On 23 December the company filed for Chapter 11 protection in Delaware and Nasdaq moved to delist it. On 27 December the executives resigned and the company disclosed that it had self-reported potential violations of law to the Department of Justice. The bankruptcy converted to Chapter 7 liquidation in March 2025. The indictment also alleges that after learning of a New York grand jury subpoena, both men deleted communications from encrypted platforms and told others to do the same, and that Chidambaran continued soliciting new investors after the fraud was public.

A civil court had already said no

Investors sued in the District of Maryland. On 20 March 2026 Judge Deborah Chasanow granted all five motions to dismiss and threw the case out. She held that the plaintiffs had not adequately pleaded falsity, had not pleaded scienter, and had not pleaded loss causation. On the central question she wrote that the suspicion that a relationship structure could facilitate revenue inflation does not mean that iLearning in fact inflated its revenues. On the defendants’ state of mind she wrote that it remained more likely that the two men had faced tricky accounting questions and relied on their independent accountant’s opinion.

She also held that the Hindenburg report could not serve as a corrective disclosure, because its authors were self-interested, disclaimed the accuracy of their own work, and drew on publicly available material that an efficient market should already have absorbed. Twenty-eight days later a grand jury in Brooklyn returned a ten-count indictment alleging that the revenue was fabricated, that the customers were shells, and that $144 million had been moved in circles to make it look otherwise. Both outcomes can be correct at once, because they answer different questions under different standards. The civil plaintiffs had to plead their way past the Private Securities Litigation Reform Act using public documents and a short seller’s inference. The government had a grand jury, subpoena power, and bank records. The gap between the two is a useful measure of how much of this kind of fraud is reachable by private litigation at all, which is to say not much of it, and not early.

Why this entry exists

By the standard used here, an entry documents a case in which artificial intelligence played a material role in committing, attempting, or enabling a crime. On a strict reading this case does not qualify. No AI capability was deployed against anyone. Logging it under the same heading as a voice-cloned ransom call would blur a distinction this record exists to preserve.

It is included because the inverse pattern is worth tracking on its own terms and is becoming common enough to count. Fraud that sells artificial intelligence rather than using it has produced a run of federal charges, and the same structural conditions produce it each time. A technology the buyer cannot evaluate, a valuation premium attached to the label, and an audit and diligence apparatus that was not built to test whether a product does what its category name implies. That is a security problem for anyone allocating capital, which is most of the readership this record is meant to serve. The appropriate long-term fix is probably a separate category rather than a tenant in investment fraud, and this entry should be read as an argument for creating one rather than as a settled placement.

What the wire coverage added

The syndicated story that carried this case to most readers was headlined around a $420 million AI business scam. That figure is the revenue iLearning claimed in 2023. It is not a loss figure, not an amount taken from investors, and not a quantity of anything anyone has recovered or is seeking. Investors lost a market capitalisation that peaked near $1.5 billion and the lenders are out some part of $60 million, but those are different numbers and the reporting did not distinguish them.

The same wire copy closed by noting that the FBI recorded more than 22,000 complaints about AI-related scams last year with losses above $893 million. The figure is real and comes from the 2025 Internet Crime Report. It has nothing to do with this case, which involves no AI-enabled scam and produced no IC3 complaints of that kind. Appending it invites the reader to file iLearningEngines under AI crime, which is precisely the classification error this entry is trying to avoid making.

What would settle it

Open. The indictment is an allegation and no trial date has been reported. The documents that would move this entry forward are the docket in the Eastern District of New York under 26-CR-97 (LDH), any superseding indictment naming the five co-conspirators, any parallel SEC enforcement action, and the Chapter 7 trustee’s findings on where the round-tripped funds ended up. Whether the auditor or the SPAC sponsor faces any consequence is also unresolved, and is the part of this case most likely to matter to anyone trying to prevent the next one.

Sources

Search used: iLearningEngines Chidambaran Naqvi indictment AI revenue fraud EDNY

Updated 2026.09.21. Tier and status definitions are in the methodology.