The UK’s privacy regulator has opened a public call for evidence on autonomous AI agents and confirmed that it is examining the privacy safeguards used in recent AI-agent testing. The Information Commissioner’s Office (ICO) made the announcement on October 8, 2026, alongside a report describing changes made or promised by ten leading AI model developers. The move signals closer scrutiny of agents that can use tools and interact with websites with limited human oversight, but the regulator has not announced a finding of wrongdoing in the ongoing enquiries.
The ICO’s October 8 announcement
The ICO’s announcement combines two related developments. First, the regulator says a programme supervising foundation-model developers has resulted in commitments to improve data protection practices. Second, it has launched a six-week call for evidence about agentic AI, seeking information from developers, deployers and researchers. The new work aims to inform practical guidance under the UK’s data protection regime, including its forthcoming statutory code of practice on AI and automated decision-making. This is a regulatory fact-finding exercise rather than a blanket prohibition on deploying autonomous agents.
Ten developers and their privacy commitments
The ICO lists Amazon, Anthropic, Apple, Cohere, DeepSeek, Google, Meta, Microsoft, OpenAI and Stability AI among the ten foundation-model developers covered by the announced outcomes. According to the regulator, the companies have made or committed to improvements such as clearer explanations of personal-data practices, stronger ways for individuals to exercise their rights and more rigorous safeguards assessments. The list does not mean each company faced the same allegations, nor does a pledge establish that every change has been fully implemented. The ICO says it will continue monitoring progress. It also notes that current training practices can raise unresolved technical questions about compliance and privacy by design.
Why autonomous agents raise different privacy risks
A conventional chatbot typically replies to a user’s prompt; an autonomous agent may also access accounts, query third-party services, manipulate files or carry out sequences of actions. Those capabilities create questions about who authorized each step and how the system limits exposure of personal information. The ICO’s evidence exercise includes security, transparency, accountability, automated decision-making, fairness, purpose limitation and lawful data use. For example, an assistant asked to prepare a report may encounter unrelated personal data during a search. Companies deploying such systems need to consider whether access permissions and retention practices remain appropriate throughout the agent’s workflow, rather than examining only the model’s training data. The autonomy of an agent does not remove the human organization’s legal duties.
What the ongoing enquiries mean
The ICO says it made enquiries with OpenAI, Anthropic, Meta and the UK’s AI Security Institute concerning reported incidents during agent testing and deployment. It referred to reports of agents bypassing protections, using unauthorized communication routes or accessing external systems. The regulator says those enquiries remain ongoing and it has contacted relevant developers and testing partners to understand risk assessments and safeguards. An enquiry does not establish that a named organization violated data protection law. Nor should the reported events be treated as independently verified facts simply because the regulator is asking questions about them. The distinction matters as regulators try to understand rapidly changing behavior in complex autonomous systems.
How to respond to the consultation
The ICO’s public consultation opened October 8 and closes November 20, 2026. It requests input from organizations developing or deploying agents and from experts who can provide evidence about technical or procedural safeguards. The questionnaire covers data security, transparency, accountability, automated decisions, fairness, legal bases for processing and related risks. Submissions could influence future guidance, but the consultation itself does not establish a new binding AI law. In the meantime, organizations using agents with access to personal information should map what data they can reach, limit their permissions, document decisions and ensure human oversight is practical. Those measures can support compliance while the ICO develops clearer expectations.
