ICO Agentic AI Call For Evidence: What UK Financial Firms Should Tell The Regulator, And Fix, Before 20 November

20 Nov 2026 - Closing date of the ICO call for evidence on agentic AI (ICO, 8 October 2026)  ·  8 - Sections in the ICO questionnaire, from data security to lawfulness of processing  ·  22A to 22D - UK GDPR articles on automated decisions, in force since 5 February 2026  ·  30 Sep 2026 - Date the ICO became the Information Commission

The ICO agentic AI call for evidence is the clearest statement yet of what the UK data protection regulator expects from firms that use AI agents. It opened on 8 October 2026 and closes on 20 November 2026. For a UK financial firm the short version is this: know every agent you run, give each one an owner and a purpose, keep records that show what it did, and keep decisions about clients with people who have the evidence and the authority to change them.

What is the ICO agentic AI call for evidence?

It is a consultation that asks organisations, developers and experts how data protection law applies to AI agents in practice. The ICO describes an AI agent as a system based on large language models that has access to tools, memory and an environment, and “has some level of autonomy in its approach to achieving a task.” The call names agent protocols such as MCP and A2A, which is a sign it is looking at how agents connect to other systems, not only at chatbots.

Announcing the call on 8 October 2026, the ICO said that “the fact AI agents act with autonomy is not an excuse for poor compliance.” The questionnaire has eight sections: about you, data security, transparency, accountability, automated decision-making, fairness and purpose limitation, lawfulness of processing, and additional questions. The ICO itself became the Information Commission on 30 September 2026, but the call is published under the ICO name.

What does the ICO want to know about AI agents?

• **Autonomy as a choice.** “The level of autonomy given to an agent should be a deliberate and carefully considered design and deployment decision.” The ICO mentions least-privilege access, restrictions on tool use and “human involvement or approval thresholds for high-risk actions”.
• **Identity.** Whether firms “identify or register an agent first” and “associate it with an owner and purpose”, including its tools, permissions and delegated authority.
• **Records.** Evidence on “structured logging measures that record AI agent identity alongside agent actions”, outcome tracking and “tamper-evident records”.
• **Tracing and audit.** Recording how personal information moves between agents and tools, and which agent took an action, with what data and with what result.
• **Purpose.** “An agent's ability to identify a new use for personal information doesn't mean that the use is automatically compatible with the original purpose.”

Why does the ICO agentic AI call matter for UK financial firms?

Financial firms handle a great deal of personal data in exactly the places agents are being tried first: onboarding files, identity checks, client emails and complaints. The section on automated decisions is the one to read closely. UK GDPR Articles 22A to 22D, in force since 5 February 2026, apply where a significant decision has no meaningful human involvement. The ICO notes that “human involvement at one stage of a workflow doesn't necessarily mean that the overall decision falls outside the ADM provisions.”

The ICO also lists what a firm should be able to reconstruct after the event: which agent was involved, what authority it was operating under, what systems or data it accessed, and whether any human involvement was meaningful. If you could not answer those four questions today for an agent that touched a client file, that is the gap to close.

What should firms put in place before 20 November?

• **A register of agents.** Every agent, including those inside vendor products, with a named owner, a purpose, the data it can see and the tools it can use.
• **Narrow permissions.** Read-only access by default. Anything that changes a record or contacts a client waits for a person.
• **Records you can trust.** A log of what each agent was asked to do, what it accessed and what it produced, kept where it cannot be quietly edited.
• **Meaningful review.** The reviewer sees the evidence, can change the outcome and has time to do so. A four-second click is not review.
• **A route for the client.** Article 22C requires firms to tell people about significant automated decisions, let them make representations, get human intervention and contest the decision.
• **Vendor answers.** Ask suppliers where agent data is processed and how actions are logged. Our article on AI agent security has six questions to start with.

How does this fit with what other UK bodies said in October?

Four days after the ICO, on 12 October 2026, the government said it must be able to “establish quickly what an agent was asked to do, what it accessed, why it acted as it did and whether the same failure could occur elsewhere.” That is almost the same list as the ICO's. Our article on AI regulation in the UK covers the government announcement, and our earlier piece on AI agents for financial firms covers the incidents behind the Bank of England's concern.

For technical teams, our playbook on agent registers and decision records turns the ICO questions into working code.

Where do AI agents fit in our own work?

Our Agentic AI London team builds agents that prepare work and hand it to a person. On client onboarding files, the agent checks each document is present, legible, in date and consistent, and lists the gaps. Decisions on taking on a client stay entirely with your people. If you want to see that on one of your own workflows, the 14-day Proof Run is the place to start. BraivIQ is a technology provider, and nothing here is legal or regulatory advice.