04 | Consumer protection
What's new?
CMA guidance on using AI Agents
04 | Consumer protection
What's new?
CMA guidance on using AI Agents
Until recently, the majority of consumer interaction with AI has been through AI tools providing consumers with direct answers, recommendation systems suggesting options, and chat‑based assistants responding to queries.
In 2026, this form of engagement changed with AI agents becoming increasingly prevalent. The efficiency and customer insight gains for businesses are obvious. For consumers, there is the potential for major benefits: significant time savings, an improved customer experience, reduced cognitive load and potentially better deals through hyper‑personalisation, price monitoring and proactive support.
A (not "the") definition of an AI agent
Can be instructed in natural language to achieve a goal autonomously, navigating some complexity in the environment, planning, coordinating, and taking actions – potentially across multiple services.
They may
- assess goals, break them into subtasks, and plan end-to-end workflows
- retrieve real-time data (including personal data) from other agents, databases and other services
- execute actions autonomously, such as making payments on behalf of the user
- store memory of past interactions to improve over time
In March 2026, following a period of research and analysis, the Competition and Markets Authority (CMA) published guidance for traders to help them comply with consumer law when using AI agents to, for example, handle customer queries, process refunds, recommend products and manage marketing campaigns.
In a nutshell, traders are advised to:
- Be transparent: label the AI agent to ensure that customers are not misled into thinking that a service is being provided by a real person.
- Design and train the AI agent with consumer protection in mind: prompts, workflows and data inputs must respect statutory and contractual rights.
- Monitor the AI agent: check, using human oversight where appropriate, for errors, bias, complaints and unintended consequences to ensure it is performing as expected and delivering the right outcomes in the real world.
- Act promptly when an issue is identified: refine prompts, workflows, training data or guardrails particularly where large numbers of consumers or vulnerable customers are affected.
CMA guidance examples
The CMA advises businesses to consider the Digital Markets, Competition and Consumers Act when designing and training the AI agent.
Using an AI agent to respond to customer service queries
- respond accurately to queries about prices, products and rights
- conduct human review of how the agent responds to queries, customer complaints or feedback
- act and correct quickly if incorrect information is given, such as regarding cancellation rights
Using an AI agent to process refund requests
- the agent's design and training must have regard to consumers’ statutory rights (for example, if they change their mind) and the terms and conditions in contracts with customers (for example, extended returns periods)
- conduct human review, for example to check that the agent’s decisions properly reflect the nature of a customer’s refund request, and any differences in time limits or amounts due.
- act and correct quickly if the agent is not consistently taking into account the extended returns policy when deciding whether a customer is entitled to return goods and claim a refund
Using an AI agent to provide, to the customer, a deal comparison service that recommends the cheapest matching deals
- results should be accurate and important information clearly and prominently displayed such as what data has been searched and any limitations in the results
- results must match the search criteria with transparency on ranking, links with suppliers, and search coverage
- conduct human review/act and correct quickly (as above)