ICO sets data protection expectations for AI agents

The UK Information Commission’s Office opened a call for evidence on agentic AI on 8 October 2026, setting out expectations for organisations managing autonomous systems that process personal data. Responses are due by 20 November 2026 and will inform future guidance and a statutory code of practice on AI and automated decision-making.

Organisations should deliberately limit agents’ autonomy through permissions, guardrails and approval thresholds proportionate to risk. The ICO highlights agent identification, named ownership and logging that allows actions and decisions to be reconstructed. Transparency may require notifications and approval steps during workflows, rather than relying solely on an explanation before processing begins.

Controller responsibilities depend on actual decisions about processing, including protocols and access controls. Automated decision-making assessments must cover the whole workflow, with meaningful human oversight rather than isolated human input. Reviewers should understand system limitations and have authority to change outcomes. Agents’ reuse of personal data also needs justification: identifying a new use does not make it compatible with the original purpose.

User instructions and service contracts do not automatically provide a lawful basis for every subsequent processing activity. Different workflow stages may require different legal grounds, while device access or storage may also require consent under electronic communications rules. The consultation accompanies enquiries to frontier AI labs and the UK’s AI Security Institute following allegations that agents bypassed protections and accessed external systems.