4 November 2026 · top ten
DRCF's 'Protecting People in an Agentic Economy': Ten Questions the Statement Doesn't Answer
Four regulators wrote one document to say they don't need to write any rules.
By end of 2027, the DRCF will publish something naming at least one gap the existing rules can't close for AI agents. The subordinate clause about "potential gaps" will become a headline. If nothing gets named, I was wrong.
The CMA, FCA, ICO and Ofcom's joint statement1, published 30 October, says current law already applies and that an agent's autonomy doesn't reduce a deployer's duty. No new rules. A promise of enforcement. These are the ten questions it doesn't answer.
10. Which regulator picks up the phone? One agent can hit data rules, financial services rules, consumer law and online safety rules in a single transaction. The CMA, FCA, ICO and Ofcom will collaborate. The statement doesn't say who goes first.
9. What counts as a deployer? A business that buys access to an agent built by one company, configured by a second and triggered by a user is deploying something. The word deployer runs through the document. It doesn't say which company in that chain it means.
8. What is the agent allowed to remember? An agent handling your money this week and your health data next week builds a cross-domain record of you. The ICO's call for evidence on AI agents closes 20 November. The statement went out before it closed. The answer is still being collected.
7. What is meaningful human oversight for a system making hundreds of calls a minute? UK GDPR Articles 22A to 22D, in force since 5 February 20262, require meaningful human involvement. They were written for single decisions. An agent running a week-long buying strategy is not a single decision.
6. Who spots agents fixing prices together? The statement flags the risk of agents optimising in ways that hurt buyers. The CMA has powers to act. It has no named programme for watching agents do it.
5. What happens when two agents, each doing nothing wrong, produce a harm together? Each supervised node follows its rules. The harm happens between them. The statement doesn't say who pays when no single link broke anything.
4. What is the user actually agreeing to? An agent acting on open instructions, across many sessions, keeping memory, is not what the consent rules were built for. The statement calls for transparency without saying what telling a user about an agent's goal would actually look like.
3. When an agent causes harm, what paper trail exists? The DRCF's March 2026 foresight paper3 flagged how hard it is to trace what an agent did across many steps. The October statement promises enforcement. It says nothing about what records a deployer must keep to make that possible.
2. What happens when the agent serves the company, not the customer? An agent a retailer deploys on a shopper's behalf serves two people. Consumer protection law was not built for a product and a service that are the same object.
1. What does the DRCF do when current law turns out not to cover this? The statement says1 the four regulators are working with government "to identify potential gaps in compliance and further regulatory challenges." That is the most important sentence in the document. It says the whole claim — that current law covers this — may not hold. It says so once, in a subordinate clause, then moves on.
The statement is not dishonest. It commits to using what exists while the gaps are still being found. That gap between commitment and gaps-being-found is what this list is about.