CMA Rules on Agentic AI and Pricing

If you've already decided an AI tool that adjusts your pricing automatically is worth using, the next question is what happens the day it makes a pricing decision that breaches consumer law. The Competition and Markets Authority has already answered that question directly.

What counts as an AI agent here

An ecommerce operator running a 20-person homeware business using an AI pricing tool that adjusts prices dynamically in response to demand, competitor pricing, or stock levels is exactly the kind of business this guidance is written for. The tool doesn't need to be a sophisticated autonomous system to count. If it's making pricing, availability, or offer decisions with limited or no human review at the point of decision, it's operating as an agent for the purposes of this guidance.

The four principles the CMA expects

The guidance is built around four principles, framed as existing consumer protection obligations applied to a new context rather than brand-new rules. Transparency: consumers must not be misled about whether they're dealing with an AI agent or about what it can and cannot do. Compliance by design: the agent must be trained and configured to respect consumers' statutory and contractual rights around pricing, refunds, cancellation, and accurate product information from the outset, not patched in afterward. Human oversight: the business must monitor the agent's actual performance, including catching errors such as hallucinated claims. Swift remediation: problems identified in the agent's behaviour must be acted on quickly, especially where the agent interacts with large numbers of customers or vulnerable consumers.

What this means for a dynamic pricing tool specifically

A pricing agent that raises prices sharply during a demand spike isn't automatically unlawful. The CMA's concern is narrower and more practical: whether the pricing change misleads the consumer about what's actually happening, whether the agent's behaviour respects existing rights (a genuine sale price must have actually been the previous price, a cancellation right can't be quietly undermined by the agent's own logic), and whether the business is actually watching for cases where the agent gets this wrong rather than treating it as fully automated and unsupervised.

The CMA has also flagged algorithmic pricing as an explicit enforcement focus for 2026 to 2027, alongside a broadened information-gathering power that lets it compel a business to produce the underlying pricing logic, training data, and decision records directly, not just a summary of pricing history.

"The algorithm did it" is not a defence

This is the single clearest message in the guidance. A business cannot point to the AI agent's own decision-making as the reason a consumer protection breach isn't the business's responsibility. The same accountability standard that applies to a staff member's actions applies to the agent's actions, and the fact that nobody at the business specifically reviewed a particular pricing decision before it happened does not remove liability for it.

The vendor's assurance is not enough on its own

A pricing tool vendor will typically say its system is built to be fair and compliant. That assurance is a starting point, not a substitute for the business's own oversight. The CMA's guidance places the compliance-by-design and human-oversight obligations on the business deploying the tool, not on the vendor that built it. A homeware retailer that has never actually reviewed what its pricing tool is doing, and is relying entirely on the vendor's marketing claim that it's compliant, is in a weaker position than one that has done even a basic periodic check of its own.

This is consistent with how the CMA has treated other forms of automated decision-making in the past: the tool is a means of implementing the business's own pricing strategy, and the business remains accountable for that strategy's effect on real customers.

What you can do about it

Three practical steps: check what your pricing tool is actually configured to do and whether it's been tested against basic consumer protection scenarios (genuine sale prices, cancellation rights, accurate availability); set up some form of ongoing human review of the agent's pricing decisions, even a periodic spot check, rather than none at all; and have a documented process for acting quickly if the agent is found to be behaving incorrectly.

This is general guidance, not a compliance assessment of any specific pricing tool. Confirm your position against the CMA's published guidance or a qualified adviser before treating any specific AI pricing deployment as low risk.

FAQ

Related reading: our AI governance by region.

Free tools: AI Tool Pricing Tracker to check current pricing across the major AI platforms | AI vs Human Cost Comparison to compare the cost of AI tools against equivalent human time.

Is dynamic pricing itself illegal under this guidance?

No. Dynamic pricing is not automatically unlawful. The CMA's concern is whether the pricing agent's behaviour misleads consumers or breaches existing consumer rights, not whether prices change in response to demand.

What penalty can a business actually face for an AI agent's pricing decision?

Under the Digital Markets, Competition and Consumers Act 2024, a business can face fines of up to 10% of worldwide turnover for breaching consumer protection rules, and may be required to compensate affected consumers.

Does this only apply to large retailers with sophisticated AI systems?

No. The guidance applies to any business using an AI agent that makes pricing, availability, or offer decisions with limited human review, regardless of the business's size or how sophisticated the underlying tool is.

What information can the CMA actually demand from a business under this guidance?

The CMA's broadened information-gathering powers can compel a business to produce underlying algorithm design documents, training data, pricing logic, and decision records directly, not just a summary of past pricing.