The growing legal risks of AI washing for businesses

With regulators already handing out punishments, business and IT leaders must be aware of the risks of misleading AI claims, writes Jonathan Armstrong.

For technology leaders, there’s often a challenge to distinguish between the real risks of AI and those conjured up by ill-informed speculation, some members of the press or sci-fi dressed up as facts. But one risk that can’t be ignored is that of AI washing - the legal risk surrounding the difference between what an AI system does vs. the organisation says it does.

Image
Description
Jonathan Armstrong is a Partner at Punter Southall Law

Claims that a product is "AI-powered", "autonomous", "secure" or capable of replacing human expertise increasingly appear in product descriptions, procurement documents, investor presentations and regulatory filings. Yet those claims may conceal extensive human intervention, depend on narrowly defined test conditions or describe capabilities that cannot be reproduced consistently in production.

This creates an assurance problem as much as a marketing one. The evidence available to engineering teams - including model evaluations, architecture documents, human-oversight procedures, incident logs and supplier assurances - may not support the broader language used elsewhere in the business.

The problem becomes more acute when organisations rely on third-party models, frequently changing services, or AI tools adopted by employees without formal adoption and approval.

The gap between a claim and the evidence supporting it is increasingly described as AI washing. It can include falsely claiming to use AI, overstating a system's autonomy or accuracy, concealing manual processes, or presenting experimental functionality as a proven operational capability.

Regulators are now testing these representations against evidence, with potential consequences including advertising restrictions, investor claims, enforcement action and criminal proceedings.

Regulators turn up the scrutiny

In September 2026, the UK's Advertising Standards Authority (ASA) challenged social media claims made by Texas-based Glow Up LLC, trading as Maxxing, about an AI-powered app designed to improve a user’s physical appearance.

The regulator found insufficient evidence that the before-and-after footage was genuine, or representative of results users could typically achieve. The ASA has indicated that the case forms part of wider scrutiny of AI product advertising across several sectors, suggesting further cases could follow.

Regulators are also examining whether businesses adequately explain AI's limitations. In April 2026, Italy's competition authority concluded investigations involving DeepSeek, Mistral AI and NOVA AI over concerns about AI-generated inaccuracies, commonly known as hallucinations. The companies agreed to improve disclosures, including warnings to users about the risk of inaccurate or misleading content. The investigations closed without findings of infringement.

In the US, regulators have taken action over exaggerated AI claims. In April 2025, the Securities and Exchange Commission (SEC) alleged that the founder of shopping app Nate raised more than $42 million by misleading investors about its AI technology, which it claimed could complete online purchases without human involvement.

According to the SEC, virtually all of the orders during the initial fundraising period were actually processed manually by contractors in the Philippines and elsewhere. The SEC brought civil proceedings, while US prosecutors also announced criminal charges.

Another case involved US legal technology company DoNotPay, which marketed its service as the 'world's first robot lawyer'. Following an investigation into claims that its AI could substitute for a human lawyer, the Federal Trade Commission (FTC) finalised a settlement with the company in 2025. DoNotPay was required to pay $193,000 in monetary relief and prohibited from advertising that its service performs like a real lawyer without sufficient supporting evidence.

What does this mean for businesses?

The risks are not confined to the technology organisations develop themselves. Businesses purchasing AI tools can still face liability if they repeat misleading claims made by suppliers.

It’s important to remember that the burden of proof is on the organisation making the claim. If your organisation makes a claim you can’t back up with evidence, trouble could result.

Conventional vendor due diligence may therefore need strengthening. Even broad assurances that AI systems are 'safe', 'secure' or 'governed' can be difficult to substantiate, particularly as models change and employees adopt shadow AI tools. Where personal data is involved, GDPR obligations around lawful, fair and transparent processing may also apply.

Six steps businesses should take

To reduce the risk of AI washing and ensure claims can be substantiated, organisations should consider the following steps:

1. Check that claims stand up to scrutiny. Any claims about AI capabilities should be supported by evidence gathered before they are made. External legal advisers may be needed to review this evidence. If AI is supplied by a third party, checks should extend down the supply chain. Publicly listed companies and those seeking investment need to take particular care.

2. Investigate complaints. Organisations should have processes in place to investigate complaints or concerns about AI claims, including those raised by whistleblowers.

3. Educate staff. Training should cover both AI literacy, which may be required under the EU AI Act, and the legal risks of making misleading claims. This should extend to senior management and board members, who often make statements to investors.

4. Test your back-up plans. Some organisations may revert to manual processes when AI systems fail or suppliers experience problems. These arrangements should be tested and properly disclosed.

5. Include AI washing in audits. Many organisations operate a three-line defence model. AI washing should be recognised as a risk within existing governance and auditing processes.

6. Map your AI systems. Maintaining an inventory of AI applications can support compliance and reduce shadow AI risks. It can also identify overlapping systems that could be consolidated, reducing costs and improving efficiency as licensing and token costs increase.