ICO Targets AI Agents: What UK Service Businesses Must Do Now

On 8 October 2026, the Information Commissioner's Office launched a six-week call for evidence on agentic AI and data protection law. In the same week, Google gave its Gemini AI model its own email address, calendar, and Drive — effectively hiring it as a digital employee. These two things together tell you exactly where AI is heading, and what UK service businesses need to do right now to stay on the right side of both the opportunity and the risk. <figure> <img src="https://images.unsplash.com/photo-1589829545856-d10d557cf95f?w=1200&q=80" alt="ICO agentic AI regulation UK service businesses — the Information Commissioner's Office launches call for evidence on data protection risks from AI agents operating across UK businesses" width="1200" height="630" loading="lazy" /> </figure> <h2>What the ICO Did on 8 October 2026</h2> <figure> <img src="https://images.unsplash.com/photo-1451187580459-43490279c0fa?w=1200&q=80" alt="ICO call for evidence agentic AI October 2026 — UK data regulator secures changes from ten major foundation model developers and seeks views from AI deployers on data protection risks" width="1200" height="800" loading="lazy" /> </figure> The ICO's intervention this month was the most significant UK AI regulatory action since UKSI 2026/425 came into force. Two things happened simultaneously. First, the ICO published the results of its year-long foundation model supervision programme. Ten major AI developers — including companies behind the models most UK service businesses use every day — made or committed to specific data protection changes. The ICO also paused engagement with xAI, the maker of Grok, after opening a formal investigation into how Grok handles personal data. That investigation is ongoing. Second, the ICO opened a six-week call for evidence asking how data protection law applies to the specific risks of agentic AI. The consultation closes on 20 November 2026. It asks developers, deployers, and experts for their views across the following areas: <ul> <li>Data security in multi-agent environments where personal data passes between multiple systems</li> <li>Transparency — how organisations disclose to individuals that an AI agent is acting on their behalf or about them</li> <li>Accountability — who is legally responsible when an AI agent makes a mistake that causes harm</li> <li>Automated decision-making and profiling — how the UK GDPR's provisions apply to agent-generated decisions</li> <li>Purpose limitation — whether data collected for one purpose can be used to train or guide an agent performing a different function</li> <li>Lawfulness of processing — the legal basis for allowing AI agents to access, process, and act on personal data</li> </ul> This is not an abstract academic consultation. The ICO is laying the groundwork for a statutory code of practice on AI and automated decision-making. The responses it receives will directly shape what becomes legally required of UK businesses that deploy AI agents. If you deploy agents that handle client or prospect data, this consultation is about you. <blockquote>The ICO does not regulate intentions. It regulates what AI agents actually do with personal data. For UK service businesses, the six-week call for evidence is not a policy exercise to watch from a distance — it is the earliest warning of what your compliance obligations will look like in 2027.</blockquote> <h2>Why This Round of Regulation Feels Different</h2> <figure> <img src="https://images.unsplash.com/photo-1504711434969-e33886168f5c?w=1200&q=80" alt="Why ICO agentic AI regulation matters more than previous rules — AI agents now act autonomously on personal data across email CRM and documents making existing guidance insufficient for the scale of deployment" width="1200" height="800" loading="lazy" /> </figure> Most previous AI regulation — the EU AI Act, UKSI 2026/425, even the ICO's earlier foundation model guidance — was written with a human in the loop as the default. A person chooses to use an AI tool. A person reviews the output. A person takes the action. Agentic AI changes that default entirely. An AI agent monitoring your email does not wait for you to open a message and decide to act on it. It reads the message, classifies it, drafts a reply, and may send it — on your behalf, using your data, without a human reviewing every step. An AI agent scanning client contracts for renewal triggers does not wait for a quarterly review meeting. It runs continuously, makes decisions, and takes actions. This matters for UK service businesses because most of you are now, or are soon to be, deployers of AI agents. The <a href="/blog/uk-ai-agent-survey-q4-2026-service-businesses">Q4 2026 survey data</a> published the week before this ICO announcement shows that 88% of UK enterprises are now deploying AI agents in some form. The ICO is not acting speculatively — it is responding to what is already happening at scale across the UK economy. The specific risk the ICO is most focused on is the multi-agent environment. When your email triage agent passes a qualified lead to your CRM enrichment agent, which passes enriched data to your proposal generation agent, personal data about your prospect has moved through three systems without that person's knowledge or consent. Under current UK GDPR guidance, each of those transfers requires a lawful basis. The ICO wants to understand how businesses are currently handling that — and, based on the evidence gathered during its foundation model supervision, most are not handling it deliberately at all. For comparison: the <a href="/blog/eu-ai-act-live-august-2026-uk-service-businesses">EU AI Act</a> went live in August and primarily concerned businesses offering AI systems to EU clients. The ICO's call for evidence is about any UK business, serving UK clients, that processes personal data using autonomous AI agents. Its scope is wider. Its effect on UK service businesses will be more immediate. <h2>Google Gives Gemini a Job Title</h2> <figure> <img src="https://images.unsplash.com/photo-1573164574048-f968d7ee9f6b?w=1200&q=80" alt="Google Gemini as digital coworker with own email calendar and Drive — AI agents becoming first-class employees in Google Workspace as major platforms reposition AI from tool to team member" width="1200" height="800" loading="lazy" /> </figure> In the same week the ICO launched its call for evidence, Google announced that Gemini agents in Workspace can now operate as full team members with their own identities. Each agent gets its own email address, its own calendar, its own Google Drive folder. It can be mentioned in documents, added to Chat spaces, and assigned to ongoing projects. Crucially, it keeps working after you close your laptop — continuing long-running tasks in the background without requiring a human to keep it open. This is not a feature update. It is a philosophical shift in how AI fits into a business. Google is not positioning Gemini as a tool you use when you remember to open it. It is positioning Gemini as a colleague you assign work to, communicate with through existing channels, and hold to account in the same way you would a junior team member. OpenAI and Microsoft are moving in exactly the same direction. OpenAI's agent products can work through Slack and Teams, managing evolving projects, updating proposals, and preparing documents. Microsoft's Autopilot operates as a digital coworker with its own identity and configurable permissions. Three of the largest software ecosystems in the world are, simultaneously, repositioning AI from assistant to employee. The regulatory implication is straightforward. An AI employee with its own email address, processing personal data about your clients, triggering actions in your systems — is precisely the kind of agent the ICO is now asking hard questions about. The technology and the regulation arrived in the same week because they needed to. For UK service businesses, this changes the calculus. The <a href="/blog/shadow-ai-risk-uk-service-businesses">shadow AI risk post</a> covered the danger of staff using unapproved tools without oversight. Google's workspace agents are the opposite of shadow AI — they are sanctioned, logged, and permission-controlled. But they still require your firm to know what they are doing with personal data, why, and on what legal basis they are doing it. Sanctioned does not mean compliant. It just means you signed up for it. <h2>Five Things UK Service Businesses Must Do Before 20 November</h2> <figure> <img src="https://images.unsplash.com/photo-1454165804606-c3d57bc86b40?w=1200&q=80" alt="Five-step action plan for UK service businesses responding to ICO agentic AI call for evidence — data flow audit agent register legal basis review transparency notices and consultation response" width="1200" height="800" loading="lazy" /> </figure> The ICO consultation closes 20 November. Even if you do not submit a formal response, the standard it sets will become the benchmark for compliance. Here are the five things to do now. 1. Audit your agent data flows. List every AI agent you currently run — email triage, lead qualification, reporting, scheduling, client onboarding. For each one, document: what personal data does it access? Where does it send data? What decisions does it make autonomously versus with human review? This is the foundation of any Data Protection Impact Assessment and it will be required for regulated sectors within months. If you built your agents using the patterns covered in the <a href="/blog/human-in-the-loop-ai-agents-uk">human-in-the-loop post</a>, your review stages are already documented. That puts you significantly ahead. 2. Build your agent register. The ICO's accountability principle requires organisations to document their processing activities under Article 30 records of processing. For agentic AI, that means maintaining a record of each agent, its purpose, the data it processes, the legal basis you are relying on, and the oversight mechanism in place. A spreadsheet is sufficient to start. The <a href="/blog/ai-governance-framework-uk-service-businesses">AI governance framework post</a> covers the three-layer approach that handles this without needing a compliance team. 3. Review your legal basis for each agent. Most UK service businesses have not formally established a legal basis for their AI agents to process client or prospect data. Legitimate interests is the most commonly applicable basis for B2B firms, but it requires a documented Legitimate Interests Assessment that weighs your interest against the individual's rights. This is not complex for most use cases — but it needs to be completed before the statutory code of practice creates an obligation to evidence it on demand. 4. Add transparency notices to client-facing processes. If any of your agents communicate with clients directly — automated follow-up emails, scheduling bots, AI-generated report delivery — those interactions require disclosure that an AI system was involved. Brief, plain-English disclosure in your engagement letter terms or email footer is sufficient for most B2B contexts. The test is whether a reasonable client would be surprised to learn an AI was involved. If the answer is yes, you need a notice. 5. Consider responding to the consultation itself. The ICO call for evidence closes 20 November 2026. Small and medium-sized businesses rarely respond to regulatory consultations — which means the resulting code of practice will be shaped almost entirely by large enterprises, foundation model developers, and consultancies. If the practical reality of running AI agents in a small UK service business is not in the evidence, the guidance that emerges will not account for it. Responses do not need to be long. The ICO has a standard submission form and a specific set of questions. An hour spent writing two pages gives the ICO evidence it would not otherwise have — and positions your firm as a thoughtful deployer rather than a passive one. <blockquote>The businesses that respond to the ICO consultation this month will help write the rules their competitors will have to follow next year. That is a direct strategic advantage available to any UK service business willing to spend an hour on it before 20 November.</blockquote> If you already have a functioning AI Operating System — multiple agents running across your core processes — the steps above are an incremental addition to what you have already built. If you are still in the pilot phase, the <a href="/blog/ai-pilot-trap-uk-production-gap">AI pilot trap post</a> is worth reading before the ICO's code of practice arrives and adds compliance obligations to agents that are already in production without governance around them. The ICO's call for evidence and Google's Gemini coworker feature both arrived in the same week for the same reason: AI agents are no longer a coming development. They are operating in UK businesses today, at scale, handling personal data, making decisions, and taking actions on behalf of people who did not individually approve each one. The question for UK service business owners is not whether to build governance around them — it is whether to build it proactively now, while the consultation is still open, or reactively once the statutory code of practice makes it a legal requirement. The Quantum Flow team works with UK service businesses to build AI Operating Systems that are both high-performing and compliant. If you want help with the data flow audit, the agent register, or reviewing your legal basis, <a href="/contact">book a free 30-minute call</a> and we will work through it with you.