What is AI-washing?

AI-washing is the use of exaggerated, misleading or unsupported claims about artificial intelligence in a company’s marketing, investor communications, product descriptions or corporate reporting.

It happens when the language used to describe a company’s AI capability does not match the reality of what the technology does.

That may include describing a product as “AI-powered”, “AI-first”, “agentic”, “autonomous” or “LLM-powered” when the role of AI is limited, heavily dependent on human oversight, or not central to the product’s operation.

AI-washing and exaggeration: when marketing claims go too far

The first and most straightforward problem is exaggeration.

A business acquires access to a foundation model, integrates it at the edges of an existing product, and announces to the market that it has transformed its operations with AI. The product itself is largely unchanged. What has changed is the label.

Describing this as an “AI-driven” or “AI-powered” offering, when the substantive work is still being done by the same processes as before, is where the exposure begins.

Agentic AI and autonomous AI claims: why language matters

The language of autonomy makes this considerably worse.

There is a meaningful difference between a tool that assists a human and one that operates independently. When a company describes its AI as “agentic” or “autonomous,” it is making a claim that will be measured against what the system actually does. If significant human oversight remains in place, if staff are routinely checking outputs, correcting errors and making the decisions that matter, then the autonomy claim is not simply aspirational. It is potentially false.

AI-powered services and hidden human labour

The second pattern involves what sits behind the AI label altogether. Some services presented as AI-powered turn out, on examination, to rely heavily on human labour to function. Where a product is marketed on the basis that it handles complex tasks at scale with minimal human involvement, but the operational reality is a workforce quietly performing much of that work, questions about misrepresentation become difficult to avoid.

Read our article on when warranties may also amount to representations for related analysis of how statements made in commercial contexts can affect liability.

AI accuracy claims and omission risk

The third issue is one of omission rather than invention.

Generative AI models are not reliable in the way that conventional software is reliable. They produce plausible-sounding output that can be factually wrong. Published data suggests that even well-regarded models carry an inherent error rate of around 3%, and this is not a temporary limitation that future development will eliminate.

It is a structural characteristic of how these systems work. Failing to disclose that in product materials, or presenting an AI tool as accurate and dependable in contexts where errors carry serious consequences, is a different kind of misrepresentation but no less serious for it.

AI-washing risk for listed companies and regulated firms

The FCA's disclosure requirements demand that information released to markets be fair, clear and not misleading. A business that describes a modest or early-stage AI project as central to its future growth strategy, without the evidence to support that framing, is making exactly the kind of statement those requirements are designed to catch.

The FCA has been clear that it wants to support innovation in AI, and that position is genuine. But it has been equally clear that AI does not sit outside the existing framework of governance, consumer protection and operational resilience obligations.

Boards that lean on AI to drive credit decisions, customer interactions or fraud detection should expect those systems to be scrutinised against those standards.

AI product claims and consumer protection risk

Where a product is sold to consumers on the strength of AI capabilities that cannot be substantiated, the broader consumer protection framework applies. The question regulators will ask is straightforward: can you demonstrate that what you told customers about this product was true?

AI and failure to prevent fraud risk

The failure to prevent fraud offence, now in force, has changed the landscape for corporate fraud liability in ways that are still being absorbed. One aspect that deserves attention in this context is the Home Office guidance accompanying the offence, which asks organisations to consider, as part of their fraud risk assessments, whether emerging technologies are creating new opportunities for fraudsters.

AI sits squarely within that question.

Deepfakes, synthetic documents, and AI-generated communications are already being used to facilitate fraud, and an organisation whose fraud prevention procedures take no account of that reality is on weak ground if the defence of reasonable prevention procedures is later tested.

AI-washing and investor claims under FSMA

Civil liability in this area does not depend on regulatory action.

Investors who acquire securities on the strength of a prospectus containing AI claims that turn out to be false or misleading may bring claims under section 90 of the Financial Services and Markets Act without needing to demonstrate that they relied on any particular statement.

Where AI has been positioned as central to a company’s growth story, and that story cannot be supported, the exposure is real.

Section 90A of the same Act extends the reach of investor claims to the broader range of published corporate information, including annual reports and market announcements, where a director acted dishonestly or with recklessness as to the truth of what was said.

In a document where AI is mentioned scores of times and framed as the engine of future performance, the argument that investors did not rely on those representations is a difficult one to make.

AI misrepresentation and breach of contract claims

Commercial counterparties have their own routes.

A business that buys an AI product on the basis of representations about its capability and autonomy, and later finds that the reality falls substantially short, has potential claims in misrepresentation and breach of contract. The more confidently the original claims were made, and the wider the gap between promise and performance, the stronger that position becomes.

How should companies reduce AI-washing legal risk?

The starting point is honesty, internally, about what the AI narrative actually says and whether it can be defended. That means going through external communications with the question not of whether they sound reasonable, but of whether they are accurate. Where they are not, they need to change.

Claims about autonomy and independence deserve particular attention. If a system described as agentic or self-directing still requires regular human intervention to produce reliable output, that needs to be reflected in how it is described.

Internal records of what oversight is in place, and how outputs are checked, are both good governance and useful evidence if questions arise later.

AI disclosures: why known limitations should be stated clearly

Known limitations should be disclosed rather than buried. In high-stakes applications, particularly those touching on financial, health or legal decisions, the fact that AI systems carry an inherent error rate is not a footnote.

It is material information that users and counterparties are entitled to have.

AI vendor contracts, warranties and insurance

Supply chain relationships matter here too. Warranties and audit rights in contracts with AI vendors provide a degree of protection where a supplier’s claims about capability prove to be inaccurate.

And insurance arrangements, particularly directors’ and officers’ cover, should be reviewed to ensure there are no exclusions that would leave the organisation exposed precisely when cover is most needed.

Read our article on price increases, contract clauses and keeping clients for related discussion of how commercial contracts should be drafted to manage changing commercial risk.

AI-washing: final word

None of this is an argument against using AI, or against communicating its genuine benefits. The concern is narrower than that. It is about the gap between what is being said and what is actually happening, and the consequences that follow when that gap is tested.

Those consequences are coming. The conditions that produce litigation, crystallised losses, disappointed investors, customers who relied on capabilities that did not exist, are assembling.

The businesses that have been careful and honest about their AI story will be in a far stronger position when that moment arrives than those that assumed the language would never be examined.

It always is, eventually.

Barnes Law’s Civil Fraud and Asset Recovery team advises clients on civil fraud disputes, deceit, misrepresentation, dishonest assistance, asset recovery and claims involving misleading commercial conduct, including disputes arising from unsupported or misleading claims about AI capability. For more information, please contact our civil fraud lawyers in London to discuss how we can support you.

Written by Barnes Law Managing Partner Yulia Barnes.