Uncategorized

The ICO and AI Agents: What UK Regulators Have Said So Far and What Marketing Teams Should Document

Marketing teams are giving AI agents access to CRMs, email lists and ad accounts. The UK Information Commissioner’s Office (ICO) has started to say what that means for data protection. This article covers what it has said, what it has not, and what a careful team should document. It is general information, not legal advice.

Key Takeaways
  • The ICO’s October 2026 statement says that AI agents acting with autonomy “is not an excuse for poor compliance.”
  • Its January 2026 Tech Futures report is early thinking, not formal guidance, and it offers no detailed mitigations.
  • We found no ICO guidance aimed specifically at marketing agents as of November 20, 2026.
  • The safest assumption is that the business using the agent stays accountable, so document purpose, access and review.

What the ICO has said

On October 8, 2026, the ICO announced data protection improvements from ten AI developers, including Anthropic, Google, Meta, Microsoft and OpenAI. In the same announcement it opened a six-week call for evidence on agentic AI. Submissions closed on November 20, 2026. The ICO said the evidence will inform future guidance and its forthcoming statutory code of practice on AI and automated decision-making.

Richard Nevinson, the ICO’s Director of Technology Regulation, said that “the fact AI agents act with autonomy is not an excuse for poor compliance.” The announcement does not set out specific duties for businesses that deploy agents. It does ask deployers how they manage the risks, which signals where scrutiny is heading. For marketers, the message is that using an agent does not move responsibility to the software vendor.

The risk areas it flags

The ICO’s Tech Futures report on agentic AI, published in January 2026, is described by Norton Rose Fulbright’s commentary of January 12, 2026 as the regulator’s “early thoughts.” The commentary says the report states that “organisations remain responsible for their use of agentic AI systems.” It also says the report does not set detailed mitigations.

Norton Rose Fulbright’s summary lists the areas the ICO raised. The right-hand column below is our reading of how each applies to marketing work, not ICO guidance.

Risk area the ICO raisedWhat it means in practiceMarketing example (our reading)
Purpose limitationPurposes may be set too broadly for open-ended agent tasksAn agent given “grow revenue” and full CRM access
Automated decisionsObligations apply where decisions have significant effectsAutomated lead scoring that blocks an offer
Transparency and rightsComplex data flows make it hard to find or correct a person’s dataAccess requests across several connected tools
SecurityAgents create new attack surfacesAn agent reading untrusted web content with CRM access
Data minimisationAgents may process more personal data than a task needsEnrichment that pulls in unneeded personal fields

What to document before an agent touches customer data

A regulator reviewing a complaint will ask what you decided and when. These steps are our recommendation for building that record. They are not ICO requirements, and a data protection adviser should confirm them for your situation.

  1. Write down the specific purpose for each agent task, narrow enough that you could explain it to a customer.
  2. List which data fields the agent can read and write, and remove everything the task does not need.
  3. Record who reviews the agent’s output before it reaches a customer, and how often.
  4. Keep the agent’s action logs, since the ICO has suggested organisations may need standalone monitoring of agent logs.
  5. Check your privacy notice still describes the processing, and confirm how you would answer an access request.
  6. Run a data protection impact assessment where the processing is likely to be high risk.

Access limits are the practical lever here. Our guide to least-privilege access for AI marketing agents covers how to apply them.

What is still unsettled

Much remains open. The ICO has not yet issued the statutory code, and we found no ICO statement written for agents running personalisation, targeting or outreach. Outbound messaging is a separate regime again, which we cover in our article on AI agent cold outreach under CAN-SPAM and UK PECR.

US businesses should not assume this falls outside their remit. UK data protection law can apply to organisations that market to people in the UK, so confirm your position with counsel. Until the ICO’s code arrives, the defensible approach is to treat agent activity as your own activity and keep records that show you did.

Frequently Asked Questions

Has the ICO published rules for AI marketing agents?

Not specifically. As of November 20, 2026, we found only its January 2026 Tech Futures report, an October 2026 call for evidence and a planned statutory code on AI and automated decision-making.

Who is responsible when an AI agent mishandles personal data?

The ICO says agent autonomy is not an excuse for poor compliance, and its report says organisations remain responsible for their use of agentic AI systems. Assume the business deploying the agent is accountable.

What should a marketing team record about an AI agent?

Record its purpose, the data it can access, who reviews its output and its action logs. That record is our recommendation and gives you something concrete to show a regulator or a client.

Clear records make agent adoption easier to defend and easier to improve. At Grow with Hafsa we help US and UK businesses set up AI marketing agents with scoped access and review steps, as part of our AI marketing and performance marketing services.