AI washing: when the sales pitch becomes a legal problem

Imagine buying software that promises to learn from your documents and answer complex questions. You pay for that capability. Later, you discover that it merely matches keywords and returns prepared answers. The supplier has changed its vocabulary, but the product has not acquired the advertised intelligence.

Now imagine an employer announcing redundancies “because of AI”. That explanation sounds decisive. It still leaves the essential question unanswered: what has actually changed in the work?

AI washing means presenting a product, service or business as more dependent on AI, or more capable because of it, than reality supports. The parallel with greenwashing is straightforward: an attractive label creates an impression the company cannot substantiate.

The phrase is fashionable. The legal problem is familiar. What matters is what the business said, what people reasonably understood and what happened because they believed it.

Advertising claims can reach the contract

A business does not need to wait for a new AI law to get into trouble over a false claim. Across the EU and EEA, the Unfair Commercial Practices Directive provides a framework against misleading consumers about a product's characteristics or expected results. Omitting important limitations can also mislead. The Commission's guidance applies these principles to digital practices. 1

Spain. The Unfair Competition Act and General Advertising Act already address misleading advertising. Claiming a capability that does not exist can lead to demands to stop or correct the advertising and, where the requirements are met, compensation. Business customers also have protection. If the promised capability was decisive in signing, an essential mistake or fraudulent inducement may justify annulment under the Civil Code. The impressive sales presentation can become evidence against its author. 2

Norway. The Marketing Control Act implements the consumer framework. Factual performance claims must have supporting evidence when the marketing takes place. Forbrukertilsynet can intervene against misleading commercial practices. Between businesses, the Act also protects good business practice and prohibits misleading representations capable of influencing demand. Deliberate deception that induces an agreement can make it non-binding under section 30 of the Contracts Act. 3

These Norwegian rules already apply. Incorporation of the EU AI Act into the EEA and Norwegian law remains pending; it is not the legal basis for this protection.

The practical point is simple: describe the capability you are selling. A buyer promised autonomous decisions should not discover that the service needs someone to perform every substantive step. As with whether an AI can enter into a binding contract, legal responsibility does not disappear inside the software.

Investors have already seen enforcement

In March 2024, the US Securities and Exchange Commission settled charges against investment advisers Delphia and Global Predictions over false and misleading statements about their use of AI. The firms agreed to pay $400,000 in combined civil penalties, without admitting or denying the findings. Those were actual enforcement outcomes, involving capabilities the firms had claimed to possess. 4

The Federal Trade Commission also launched Operation AI Comply in 2024. Its actions included allegations against DoNotPay over claims that its service could replace a lawyer's expertise, alongside cases involving supposedly AI-powered business opportunities. 4

European investment firms face their own duty to give clients information that is fair, clear and not misleading. ESMA has expressly applied that duty to explanations of AI use. Breaching investor-information rules can carry sanctions. A confident pitch about technology is still a representation to someone deciding where to put money. 4

Saying “AI” does not justify a dismissal

Challenger, Gray & Christmas recorded 54,836 announced job cuts in 2025 in which US employers cited AI. That wording matters: it records the explanation attached to announcements. The debate covered by Harvard Gazette asks how those announcements relate to changes in work actually observed. 5

For an employer, the legal question is concrete. Which tasks will change? What technology supports that change? Why does it make the position redundant? Calling a reorganisation an AI transformation does not answer those questions.

Spain. For an individual dismissal based on economic, technical, organisational or production grounds, the employer must establish the cause stated in the dismissal letter. If it cannot, the dismissal is unfair, with reinstatement or statutory compensation as the usual consequences. A genuine technical change can support a dismissal. An unsubstantiated reference to AI cannot supply the missing facts. Collective redundancies also require their own procedure. 2

Norway. A dismissal must be objectively justified under the Working Environment Act. Before deciding, the employer must, as far as practicable, discuss the grounds and any selection between employees with the employee and their representative. Redundancy involves objectively justified selection, consideration of suitable alternative work, and balancing the business's needs against the employee's disadvantages. The employer must substantiate the factual basis. An unjustified dismissal can be declared invalid and lead to compensation. 3

If a company uses the language of innovation to dress up a decision it cannot justify, the language will not rescue it. The evidence has to explain the decision.

South Korea brings the issue back into focus

On 22 October 2026, South Korea's Fair Trade Commission opened consultation on draft guidelines for AI-related labelling and advertising, with comments due by 11 November. Its announcement addresses claims of AI use where none exists, as well as misleading omissions about scope, operating conditions and performance limits. 5

The draft is non-binding guidance explaining existing advertising principles through examples. That is precisely why it matters to this discussion. Authorities can make their expectations more explicit without creating a new offence called “AI washing”. A familiar duty remains at the centre: make claims you can support.

Sell what the system actually does

Before announcing an AI feature, I would want three plain answers: what it does, what it cannot do, and what evidence supports the claim. Those answers should survive beyond the marketing meeting. They should match the demonstration, the contract and the service the customer receives.

If people review the output, explain their role. If a claimed improvement comes from a limited trial, describe that trial accurately. Keep the evidence before publishing the claim. A promise that needs an entirely different explanation once a customer complains was probably badly drafted from the start.

There is plenty to sell in a tool that prepares documents, organises information or helps staff respond faster. The argument for putting AI in the back office starts with useful work. Businesses do not need imaginary autonomy to explain a real improvement.

My view is that AI washing becomes dangerous when the label replaces the explanation. Customers, investors and employees are entitled to the facts relevant to the decision in front of them.

Say what your AI does. Be ready to demonstrate it. And remember: saying “AI” is not a legal ground for dismissal.