When AI gets it wrong: Could a mistake made by AI result in a business dispute? 

AI agents are moving further into the customer journey. They can answer questions, recommend products, process refunds and increasingly take actions on a customer's behalf.

Recent reporting on AI shopping agents shows how quickly this is developing. Reuters reported that major banks were warning about risks including fraud, scams and data privacy as AI agents become more involved in online shopping.  For businesses, the issue is not simply whether an AI agent can make a mistake. It is what happens when a customer relies on that mistake.

That raises an important commercial question: when an AI generated mistake causes a customer to suffer a loss, where does responsibility sit?

When a mistake by an AI agent becomes a business’s problem

The Competition and Markets Authority (CMA) has already addressed part of this question. The Governments guidance on using AI agents says that businesses remain responsible for complying with consumer law when they use AI to deal with customers. The fact that an AI agent is supplied by a third party does not transfer that responsibility elsewhere.

The CMA gives the example of an AI agent repeatedly giving customers the wrong information about product features or cancellation rights. Businesses are expected to monitor how their AI agents perform and act quickly when problems are identified.

That could become particularly important where a customer relies on information provided by an AI agent and acts on it.

An inaccurate answer does not automatically mean there is a legal claim. But if a customer relies on that information, enters into a transaction and subsequently suffers a loss, the mistake could become more than a customer service problem. It could become a dispute.

Then there is the AI provider

There may be another question behind the customer dispute. If the business did not build the AI agent itself, could it recover its losses from the provider? That will depend on the contractual relationship.

Businesses using third-party AI agents should consider what their agreements say about liability, indemnities, exclusions and limitations of liability. They should also understand what the provider is responsible for, what the business is expected to monitor and what happens if the system changes.

These provisions may not receive much attention when everything is working as expected. They could become much more important when something goes wrong. There is also an evidential issue.

If a customer says an AI agent promised something, how does the business establish exactly what was said? Was the interaction recorded? What information did the agent have at the time? Which version of the model was being used? Had the system or its instructions changed since the interaction?

Those questions could matter if the dispute develops into a legal claim.

None of this means businesses should avoid using AI. The CMA recognises the potential benefits of AI agents, while emphasising the importance of testing, monitoring and human oversight.

For businesses, the practical question is whether those protections are being considered alongside the commercial and contractual risks.

Before allowing an AI agent to interact directly with customers, businesses should consider:

• What can the AI agent say or do? Set clear limits around customer facing communications and decisions.
• What happens when it gets something wrong? Have a process for identifying, correcting and escalating errors.
• What records are kept? Consider how the business would establish what happened if a customer later challenged an AI-generated statement.
• What does the AI contract say? Check liability provisions, indemnities, exclusions and obligations on the provider.
• When does a human need to step in? Some decisions may be too commercially or legally significant to leave entirely to an automated system.

The business case for AI is increasingly clear. The more difficult question is what happens when the technology fails.

An AI agent can make a mistake in seconds. Resolving the consequences may take considerably longer.

As AI takes a greater role in customer relationships and transactions, businesses should not only ask what the technology can do for them. They should also ask what happens when a customer relies on it, something goes wrong and the relationship turns into a dispute. The AI may have delivered the answer. But if that answer causes a dispute, the business may still be the one left dealing with the consequences.

Ian Carson, Head of Dispute Resolution at Harper James, comments:

Businesses should not assume that an AI generated statement sits outside the legal relationship with their customer. If a customer has relied on information provided by an AI agent acting on the business’s behalf, the circumstances of that interaction could become highly relevant if a dispute arises. Businesses need to understand what their AI agent is authorised to say and do, and what protections are in place when it gets something wrong.