The report
The UK Information Commissioner's Office opened a consultation on agentic AI on October 8, 2026, asking how existing data-protection rules should apply when AI systems act through tools, process personal information and make or influence decisions. The official call for evidence is open until November 20. It is a request for input before final guidance, not a new statutory rule or a ruling that the platforms discussed in security research have broken the law.
The ICO asks for evidence on data security, transparency, accountability, automated decision making, fairness and purpose limitation, and lawful processing. A conventional chat interface may reveal only the final answer; an agent can additionally query stored conversations, read records through a connector or initiate an action in another service. The practical legal and technical question is who decides which data is necessary and who enforces the action boundary after the model proposes the next step.
Two concrete boundaries worth examining
LangChain's MongoDB session advisory illustrates how an application can fail to segregate stored chat histories when it trusts a structured user-supplied session identifier. AgentCorruption's controlled IAM test illustrates how a cloud-agent role can reach other sessions and memories when its permissions are too broad. These investigations are independent of the ICO consultation and are not allegations that the regulator investigated either vendor. They show the types of data-access boundaries that effective agent governance must be able to explain and test.
Responses to the consultation could help clarify what evidence developers and operators need to preserve about agent data sources, tools, permissions and actions. The ICO has not yet published final agentic-AI guidance through this consultation. Until then, the established obligations and each product's actual implementation remain the relevant basis for compliance assessment, rather than hypothetical new duties inferred from the consultation questions.