AI-Washing: When the “AI” Turns Out to Be a Wrapper

The US securities regulator brought its first AI-washing cases in 2024. Founders have since faced criminal charges. For investors, customers and acquirers in Singapore, the private claims have barely started.

What is AI-washing?

If you have invested in an AI company, or your business has recently bought an AI tool, pay close attention. AI-washing is the practice of overstating what artificial intelligence a product actually uses, or how well it works, in order to win over investors or customers.

Picture a scenario where a company advertises its product as one with an automation rate of 95%, suggesting an efficient, AI-powered process which reduces the risk of human error – but in reality, the product is unautomated and simply operated by human contractors on the back end. In other cases, a company advertises a bespoke AI tool developed by industry professionals, trained on sector-specific data sources and running on a proprietary model – but in reality, the tool simply runs on OpenAI’s GPT models (the models behind ChatGPT) with no value added by the company. These are prime examples of AI-washing. Since 2024, regulators in the USA have brought a steady stream of AI-washing cases, and in some cases founders have also been criminally charged.

If you relied on such false statements in making an investment or purchase, Singapore law may give you a claim in misrepresentation, in the tort of deceit, under your contract and, in some cases, under the Securities and Futures Act 2001 (the “SFA”).

The pattern

The cases now have a recognisable shape. In most, the technology behind the product was described as “proprietary” but was in truth a third-party model, a manual process, or both. Investors and purchasers often discovered the truth only at the down round or when the company went into insolvency.

Some prominent examples:

Company	What was claimed	What was found or alleged	Outcome
Nate (US shopping app)	An AI that completed online purchases at a 93 to 97 per cent automation rate. Raised more than US$40 million from investors including Coatue, Forerunner and Renegade.	According to US prosecutors and the SEC, the automation rate was effectively zero. Purchases were made by contractors in the Philippines and Romania.	Founder indicted for securities and wire fraud, and sued by the SEC (April 2025). He has pleaded not guilty; the case is pending.
Joonko (US recruitment platform)	A product that carried out candidate matching driven by “seven different AI algorithms”.	The SEC alleged the technology was far less advanced than claimed. Prosecutors say customer and revenue figures were also falsified, backed by forged bank statements and purchase orders.	Founder charged by the SEC and US prosecutors (June 2024); pleaded guilty to securities fraud (September 2026). About US$27 million was raised from investors.
Presto Automation (US-listed restaurant technology)	Drive-through voice ordering technology described as its own, which “eliminat[es] human order taking”.	For about a year, the product ran on a third party’s speech-recognition technology. Even Presto’s own version needed human intervention in most orders.	SEC issued a cease-and-desist order (January 2025). No penalty, in light of the company’s financial condition, remediation and cooperation.
Builder.ai (UK/US app-building platform)	An “AI-powered” platform for building software. Raised about US$450 million from investors including Microsoft, the Qatar Investment Authority and SoftBank.	Long-standing allegations that much of the “AI” work was done by human engineers, and reports that revenue had been overstated.	Entered insolvency in May 2025. US authorities are reported to be investigating.

The warning signs

Investors in, and buyers of, AI products should watch for the following red flags:

Cost of revenue or contractor spend that grows with every new customer, when automation should have kept it flat.
Automation or accuracy figures that come without logs, test data or a methodology.
Demonstrations given only in controlled conditions, and never on your own data.
A refusal to name the underlying model or explain the architecture, merely justified as “proprietary”.
Payments made to outsourcing firms without a proper description, or with vague descriptions such as “data labelling” or “quality assurance”.
Departures from the engineering team, late audited accounts, or revenue that comes largely from related parties.

Who has a claim?

Investors

Funds, family offices and angel investors who subscribed for shares on the strength of what they were told about the technology. Depending on the facts, their claims may lie against the company and against any founder who made, authorised or assumed responsibility for the false statements.

Businesses that bought an AI product

A company that signed a multi-year licence for an “autonomous” claims-processing, KYC or customer-service tool, and then found that the AI was a team of offshore staff or a thin layer over a public model, may likewise have a claim in misrepresentation.

A purchaser may also have potential claims in contract, for example, claims concerning service levels, accuracy warranties and termination rights as stipulated in its agreement with the seller. Whether the purchaser can recover fees already paid will depend on the contractual terms, the available remedies and the facts. There may also be a data problem – if customers’ information was being handled by people it never knew existed and in countries it never approved, potential contractual claims may arise, and this may involve breaches of the Personal Data Protection Act 2012.

Acquirers

Buyers who paid a premium for a target’s “proprietary AI” may have the most money at stake, and often the widest range of claims. Warranties contained in the sale and purchase agreement on the target’s technology, intellectual property and accounts may be breached if the “AI” turns out to be a third-party model or a team of offshore staff. Where the warranties are capped or have been disclosed against, a claim in misrepresentation or deceit against the sellers may sit alongside them. The case study involving British software company Autonomy shows that these claims can succeed against individuals even after extensive due diligence has been conducted on the target. American technology company Hewlett-Packard (HP) bought Autonomy for approximately US$11 billion in 2011, subsequently alleged that Autonomy’s revenues had been fraudulently inflated, and sued its founder and finance director personally. Damages of around £700 million were assessed in July 2025. Autonomy was an accounting case rather than an AI one, but AI-washing raises the same question in a new form: what was the buyer told about the business, and was it true?

What claims can be brought in the context of AI-washing?

In Singapore, the most likely claim in the context of AI-washing is under section 2(1) of the Misrepresentation Act 1967. An investor or purchaser must prove that the other party to the contract made a false statement of fact that induced them to enter into the contract, and that they suffered loss as a result. The burden then shifts to the party who made the statement to prove that it had reasonable grounds to believe, and did believe, up to the time the contract was made, that the statement was true. Where a product was described as “proprietary AI” and plainly was not, that burden is close to impossible to discharge, particularly if the description was unqualified and unsupported by contemporaneous records. The claimant can recover the loss which it suffered by entering into the transaction.

Where the party making the statement knew it was false, or did not care whether it was true, the claim also lies in the tort of deceit or fraudulent misrepresentation. That adds three key considerations in favour of the claimant:

non-reliance clauses (i.e., clauses under which an investor or purchaser confirms that it did not rely on any statements made outside of the written contract before entering the contract) cannot exclude liability for fraud;
the limitation period (i.e., the period of time which a claimant has to commence a claim) starts running only from when the fraud was, or could reasonably have been, discovered; and
the founders’ personal assets come into play, and recovery is not restricted to the assets of the company.

Where a false statement of fact was made to induce a subscription for shares, section 199 of the SFA may also be engaged. The statute prohibits false or misleading statements likely to induce others to subscribe for or buy securities, and a contravention is a criminal offence.

A claim for breach of contract under the subscription agreement or the licence may also be viable, but such claims may be capped, limiting the amount that a claimant may be able to recover. Further, contractual warranties may not necessarily say anything about the architecture of the product, and as such whether or not such a claim would succeed would depend heavily on the specific contents of the contractual warranties.

What counts as a misrepresentation?

Suppose a company describes its product as “AI-powered” – can an investor or purchaser argue that such a label constitutes a misrepresentation that forms the basis of a claim? Not necessarily. Words such as “revolutionary”, “AI-first” and “intelligent” may amount to marketing puffery. On their own, they may be too vague to constitute statements of fact.

However, the statements that actually persuade an investor or customer to put money down tend to be specific and verifiable. A seller may say, for example, that its product:

runs on a proprietary model;
is trained on the company’s own data;
is fully automated, with no human in the loop;
achieves a stated automation or accuracy rate;
has a stated number of customers; or
generates a stated amount of recurring revenue.

These are statements of present fact. A wrapper around someone else’s model, sold as proprietary technology, is plainly a misstatement of fact. Forecasts and statements about the future are not statements of fact in themselves, but they may carry an implied representation that the maker honestly believes them and has reasonable grounds for them.

As a defence, founders engaging in AI-washing may argue that the investor or purchaser knew all along that the product ran on a third-party model (e.g., OpenAI) underneath before entering the agreement. If a founder states upfront that the product runs on a third-party model, such disclosure may defeat an allegation that the product was misrepresented as running on a proprietary model. The effect of the disclosure will depend on its clarity, timing and consistency with the other materials provided. In turn, this depends heavily on the documents presented to the investor or customer such as the pitch deck, data-room Q&A, investor updates and the recording of the demonstration.

The documents can cut both ways. If a data room or subscription agreement told the truth where the pitch deck did not, a Singapore court may find that the deck did not in fact induce the investment. In a recent High Court decision, sophisticated parties were taken to have read contract documents that corrected the alleged representations. Equally, answers given in due diligence are representations in their own right, and are often more precise than the deck. In one English case, the High Court held founders liable in deceit for false answers in a founders’ questionnaire.

Getting the money back

If you invested in or bought a product on the strength of a representation about the technology that you now doubt, the time to look at the documents is before the next round, renewal or insolvency, not after. Speed matters, particularly where there is a risk that assets may be disposed of. Where you have a good arguable case and there is a real risk that the defendant will dissipate its assets to frustrate enforcement of a judgment, the court can grant a freezing injunction over assets in Singapore or worldwide. The court may also order a bank or other non-party to produce documents and information, including before proceedings are commenced, so that you can identify possible parties to a claim or trace your property, subject to the necessary requirements being met. These applications can be made urgently and, where secrecy is needed, without notice to the other side.

These disputes are rarely confined to one country. An AI-washing structure may involve a company incorporated in a foreign jurisdiction such as the Cayman Islands, founders based in the UK, money from a Singapore family office, and a shareholders’ agreement choosing Singapore law as the governing law and SIAC arbitration as the dispute resolution forum. While the forum is usually fixed by the investment documents, the assets may be anywhere, which is why worldwide freezing orders and early tracing matter.

If it is too late to freeze assets, recovery from the founders themselves can be difficult. Their wealth may be tied up in shares in the company, which are now worthless, and investors’ money was paid to the company, not to the founders themselves. The position may differ where founders received sale proceeds, substantial payments or transferred assets out of reach, and that is where an early freezing injunction may pay for itself and materially improve the prospects of recovery.

Delay can also reduce what you recover. In a recent representative action brought by cryptocurrency investors in the Singapore International Commercial Court over the collapse of the TerraUSD “algorithmic stablecoin”, the defendants conceded that key representations about how the technology worked were false. Even so, the investors recovered only a small fraction of what they claimed: losses suffered after the misrepresentation ceased to be operative were not recoverable. Once you know, or ought to know, the truth, continuing to hold the investment is at your own risk.

Further, if the company goes into liquidation or judicial management (which is often the case where the company is no longer able to pay its debts as and when they fall due), some claims against founders and advisers will belong to the company, and fall to be pursued by the liquidator or judicial manager rather than by individual investors. It is critical to engage with these insolvency professionals early, including on who will bring those claims and how they will be funded.

If your business involves AI products or tools

The same principles apply in reverse. If your company describes its product as AI-powered in pitch decks, on its website, in listing documents or in fund materials, the time to check those statements is before an investor, customer or regulator does. Every material statement should be backed by a substantiation file: what the system actually does, what the automation and accuracy figures are measured against, and which third-party models (if any) sit underneath.

A false or misleading statement likely to induce a subscription for securities can be an offence under section 199 of the SFA. For consumer-facing products, the Consumer Protection (Fair Trading) Act 2003 treats representing that goods or services have performance characteristics, uses or benefits that they do not have as an unfair practice. Directors who approve the statements may also face personal liability for breaches of directors’ duties.

Summary

What is AI-washing?

AI-washing is overstating what artificial intelligence a product uses, or how well it works, to win investment or customers. Unlike marketing puffery, it involves specific claims that can be checked, such as automation rates or the use of a proprietary model.

Is AI-washing illegal in Singapore?

There is no statute aimed specifically at AI-washing. However, false statements about a product’s AI can give rise to civil claims in misrepresentation, deceit and contract and, in some cases, to offences under the SFA or findings of unfair practice under the Consumer Protection (Fair Trading) Act.

Can I sue if a startup I invested in overstated its AI?

Potentially, if you can identify a specific false statement of fact that induced your investment. Claims can be brought against the company and, where the elements of deceit are established, against any founder who is personally responsible for the fraudulent representation. The documents you received before investing will usually decide the case.

What if a vendor we bought an AI product from overstated what it does?

You may have a claim in misrepresentation as well as contractual remedies. Depending on the facts and the agreement, those remedies may include termination, rescission, damages or recovery of some or all fees paid. Check the warranties and data-handling terms in your agreement.

How long do I have to bring a claim?

Generally six years for claims in misrepresentation and contract. Where fraud is involved, time runs only from when the fraud was, or could reasonably have been, discovered. Waiting is rarely wise, however: assets tend to move long before the limitation period expires and you may face difficulties obtaining recovery.

How we can help

The team at 26 Gambit Chambers has experience acting for a wide range of investors, businesses and acquirers in disputes across different sectors, and provides corporate advisory to companies and boards, including on transactions and directors’ duties. If you invested in, bought or acquired a business on the strength of a representation about its technology that you now doubt, or you want your own statements checked before someone else does, the time to act is now.

This article is for general information only and does not constitute legal advice.