CMA warns AI agents may push consumers toward worse deals

TL;DR:

• The UK’s Competition and Markets Authority has published a report warning that agentic AI assistants could manipulate consumer choices, push higher-priced products, and quietly serve the interests of the companies behind them.
• The watchdog flags risks including AI hallucinations leading to costly autonomous errors, personalisation enabling harder-to-detect dark patterns, and consumers gradually losing the ability to scrutinise automated decisions.
• The CMA is not proposing new rules, arguing that existing consumer protection law already applies whether a decision is made by a human or a machine.

The CMA has fired an early warning shot at the agentic AI industry. In a report published Monday, the UK competition regulator examined AI systems designed to handle tasks like comparing services, booking travel, switching providers, and managing subscriptions on a consumer’s behalf.

The central concern: these agents may not be “faithful servants” to the people using them. An AI assistant that appears to be hunting for the best deal could instead surface products generating more revenue for the platform operating it.

Dark patterns at scale

The CMA argues that personalisation, often marketed as a feature, could make manipulation harder to spot. When every user receives different recommendations and prices based on behavioural profiles, detecting unfair steering becomes difficult for both individuals and regulators.

Highly adaptive agents could amplify the manipulative interface tricks known as dark patterns, particularly if the underlying systems are optimised for engagement or conversions rather than consumer benefit. The report draws a clear distinction between an AI that advises and one that acts: an incorrect chatbot response is annoying, but an autonomous agent cancelling a service or switching a contract based on flawed reasoning could be expensive.

Reliability and oversight gaps

The watchdog also highlights the reliability problem. Current AI models remain prone to errors and hallucinations, and those mistakes carry higher stakes when agents take action rather than offer suggestions. If agents use complex multi-step reasoning that consumers cannot inspect, unfair outcomes become harder to challenge under existing frameworks.

There is also the question of attention. As people delegate more decisions to automated assistants, the CMA warns of growing “over-reliance” where consumers lose the habit of checking what their agents are doing.

Despite the lengthy risk assessment, the regulator is not proposing new legislation. Its position is that current consumer protection law applies regardless of whether a human or machine makes the decision. Companies operating AI agents that steer consumers toward misleading deals will still face enforcement.

The timing of this report is worth noting. It arrives the same week a US court blocked Perplexity’s AI shopping agent from accessing Amazon, suggesting regulators on both sides of the Atlantic are moving quickly to define the boundaries of agentic commerce.