Agentic AI needs a data-action map before privacy guidance arrives
The UK ICO is gathering evidence on agentic AI until 20 November. Deployers should map every data-reading and data-changing action now instead of waiting for final guidance.
What happened
On 8 October the UK Information Commissioner’s Office opened a six-week call for evidence on agentic-AI data-protection risks and said enquiries into reported agent-testing incidents were continuing.
Why it matters
The call covers security, transparency, accountability, automated decisions, fairness, purpose limitation and lawful processing—issues that can attach to each tool action, not just the model prompt.
The Information Commissioner’s Office announcement says its agentic-AI call for evidence runs for six weeks and closes on 20 November 2026. It seeks operational evidence on security, transparency, accountability, automated decision-making, fairness, purpose limitation and lawful processing. The ICO also says it has contacted developers and testing partners about reported agent-testing incidents; those enquiries are ongoing.
The same update reports that ten foundation-model developers made or committed to data-protection changes after supervisory work. That wording matters: some measures are commitments rather than completed controls. Independent regulatory analysis likewise separates completed changes from future commitments and notes that the agent call will inform later guidance and a statutory code. The call itself is evidence gathering, not a final interpretation of every legal boundary.
Map data to actions, not just systems
Create one row for each action an agent can take: data read, data written, purpose, lawful basis asserted, recipient, retention, user notice, human decision point, reversal path and evidence record. Include actions initiated through tools, memory, background execution and delegated agents. Flag any row whose purpose or recipient changes after the initial user instruction.
Use the map for an internal challenge session and, where relevant, a bounded response to the call.
Test one request that changes purpose mid-run and one request delegated to a service with a different retention rule. Record whether the agent pauses, whether the person sees the change and whether the resulting data action can be undone. Do not label the inventory “ICO compliant”. The immediate test is whether an owner can explain and reverse a material data action before final guidance arrives. Connect the control to the human-in-the-loop AI skill, with a separate privacy and legal review.