The ICO agentic AI call for evidence has one month left to run. The UK's data protection regulator wants developers, companies that deploy AI agents and independent experts to explain how they handle the privacy risks of software that can act on its own. Responses are due by 20 November 2026.
The consultation sits alongside a second piece of work the Information Commissioner's Office published on 8 October: a report on what it extracted from ten of the largest foundation model developers after a two-year supervision programme. Read together, the two documents show where UK enforcement on AI is heading, and the agents are now at the front of the queue.
What the ICO agentic AI call for evidence asks
The ICO describes AI agents as systems built on foundation models that can complete tasks, use tools and interact with websites, "often with limited human oversight". Its view is that the data protection risks change as these systems gain autonomy, and it wants evidence on how organisations are dealing with that today.
The call for evidence runs for six weeks and covers these areas:
- security of agent systems and the data they touch
- transparency towards the people whose data an agent processes
- accountability when an agent acts without a human approving each step
- automated decision-making and the rights attached to it under UK GDPR
- fairness of outcomes
- the lawful basis for using personal data in agent workflows
The regulator says the answers will feed into future guidance and into its forthcoming statutory code of practice on AI and automated decision-making. That code matters more than ordinary guidance, because a statutory code is something the ICO and the courts take into account when judging whether an organisation complied with the law.
Enquiries already under way
The ICO has not waited for the consultation to close. It says it has made enquiries with OpenAI, Anthropic, Meta and the UK AI Security Institute about recent testing and deployment of agentic AI. According to the regulator, reports indicate that some agents bypassed protections, used communication channels they were not authorised to use and reached external systems such as Hugging Face.
It has also contacted several developers and their testing partners to ask what risk assessments and safeguards were in place. Those enquiries are continuing, and the ICO has not said whether any could lead to formal action.
Richard Nevinson, the ICO's Director of Technology Regulation, put the regulator's position plainly: "the fact AI agents act with autonomy is not an excuse for poor compliance."
What ten AI developers agreed to change
The second document is a report on the foundation model supervision programme the ICO set up in 2025 under its AI and Biometrics Strategy. The programme ran for two years and covered 11 developers chosen by likelihood of non-compliance, UK market share and use of higher-risk training data.
Ten of them made changes or committed to them: Amazon, Anthropic, Apple, Cohere, DeepSeek, Google, Meta, Microsoft, OpenAI and Stability AI. The ICO says the changes fall into three groups:
- clearer transparency information about how personal data is used
- stronger ways for people to exercise their data protection rights
- tougher assessments of the safeguards developers rely on
The eleventh developer was xAI. The ICO paused its engagement with the company after opening a formal investigation into the Grok AI system, and that investigation is still open.
The report also sets out the regulator's position on two questions that have dogged generative AI since it arrived: how special category data, such as health or ethnicity information, can lawfully be used, and whether a trained model can itself contain personal data. The ICO also concedes that current training practices create technical problems for UK data protection law and for "data protection by design", and says it is raising those problems with the government.
Who should respond, and why it matters outside the UK
UK GDPR applies to organisations outside the country when they offer services to people in the UK or monitor their behaviour. A start-up in Lagos or Nairobi that sells an agent-based customer service product to British clients is inside that scope, as is any company that lets agents browse, book or buy on behalf of UK users.
Practical groups with a reason to reply include:
- Agent developers, who can explain what controls exist on tool use, memory and external access
- Deployers, such as banks, retailers and public bodies, who decide what an agent may do with customer data
- Auditors and researchers, who have seen agents behave in ways their builders did not expect
The question of who controls an agent's actions is already shaping commercial decisions. Earlier this week we reported how Schwab and JPMorgan blocked outside AI agents from moving customer money, a sign that firms are drawing their own lines before regulators do.
The wider ICO picture
Agents are one of several AI priorities at the regulator. The ICO says it is also researching public concerns about increasingly personalised consumer AI, including general-purpose chatbots and companion or role-play chatbots, and is talking to companies about transparency and privacy in those products. Earlier in 2026 it published its first thinking on agents in a Tech Futures report titled "AI'll get that!".
There is also an institutional change in the background. Under the Data (Use and Access) Act 2025, the regulator's legal entity is now the Information Commission, overseen by a new board, rather than a single Information Commissioner. The ICO name remains in use.
What happens next
After 20 November the ICO will analyse responses and use them to draft guidance and the statutory code on AI and automated decision-making. No publication date has been given for the code. The enquiries into OpenAI, Anthropic, Meta and the AI Security Institute will run in parallel, and the xAI investigation remains separate.
For companies building or buying agents, the message from the regulator is that the usual data protection duties apply in full, however much of the work the software does by itself. Submissions can be made through the ICO's call for evidence page.
