AI agents – tools that don't just answer a question but take actions on your behalf, such as following up a lead, booking a meeting or handling a customer query end to end – are moving quickly from novelty to normal. If you are piloting one in sales or customer service, a statement from the Information Commissioner's Office this month sets out the ground rules, and the headline is blunt: autonomy is no excuse.
What the ICO said
On 8 October 2026 the ICO announced it had secured data-protection changes from ten major AI developers, including Google, Microsoft, OpenAI, Anthropic and Meta, covering clearer transparency, better ways for people to exercise their rights, and more rigorous safeguards. It also opened a call for evidence on the data-protection risks of “agentic AI”, which closes on 20 November 2026, and said the evidence will feed future guidance and a statutory code of practice on AI and automated decision-making.
The line that matters most for businesses: in the ICO's words, “the fact AI agents act with autonomy is not an excuse for poor compliance.” If an agent handles personal data on your behalf, the responsibility for doing that lawfully is yours, not the tool's.
Why this lands on established businesses, not just AI labs
The developers make the models. But the business that points an agent at its customer list, its inbox or its CRM is the one deciding how real people's personal data gets used. That makes you the data controller, with the duties that come with it – regardless of how clever or autonomous the tool is.
The risk is not theoretical. The ICO noted reports of agents bypassing protections, using unauthorised channels and accessing external systems. An agent that quietly emails the wrong list, stores data somewhere it should not, or makes a decision about someone with no human check is your compliance problem to answer for.
A practical checklist before you deploy an AI agent
- Know what data it touches. Map exactly what personal data the agent can read, write and send, and to where. If you cannot describe this, you are not ready to switch it on.
- Keep a human in the loop for decisions about people. Automated decisions that significantly affect someone – who gets a quote, who is turned away – carry specific obligations. Build in a human check.
- Be transparent. If customers are interacting with or being processed by an AI agent, they are entitled to understand how their information is handled. Say so plainly.
- Limit its reach. Give the agent the minimum access it needs, not the keys to everything. Restrict which systems and channels it can use.
- Log and review. Keep a record of what the agent did, and review it. “The AI did it” is not a defence.
- Check your contracts. Make sure your agreement with the AI provider covers how they handle data on your behalf.
This is the responsible way to get the efficiency gains without the exposure – the same balance we describe in using AI to run the business better.
Frequently asked questions
Does using a well-known AI provider make us compliant?
No. The provider's commitments help, but when you deploy an agent against your own data you are the data controller and remain responsible for lawful use. Choosing a reputable provider is a start, not a substitute for your own controls.
Can an AI agent make decisions about customers on its own?
Where a decision significantly affects a person, UK data protection law places specific requirements on solely automated decisions, including rights for the individual. The safe default is to keep a human involved in those decisions.
Should we respond to the ICO's call for evidence?
If you are actively deploying agents, contributing before it closes on 20 November 2026 is a way to shape the guidance. At minimum, follow it, because it signals where the forthcoming statutory code is heading.
Automate with the guardrails in place
We help businesses put AI and automation to work in sales and operations without creating new risks. See our AI, automation and process optimisation service, or get in touch for a straight conversation about what's worth automating and what isn't.
Sources
- ICO: ICO secures changes from leading AI developers as scrutiny extends to AI agents (8 October 2026)
- TLT: The ICO's agentic AI call for evidence – what you need to know
This article is general information, not legal advice.
