The UK Information Commissioner's Office opened a call for evidence on 8 October about how data-protection law should apply to agentic AI. The regulator is asking developers, deployers and other experts about systems that can complete tasks, use tools and interact with websites with limited human oversight. For charities, the immediate task is not to assume that autonomy transfers responsibility: before an agent touches supporter, staff or service-user data, the organisation should be able to show what it may do, why the processing is lawful and how a person can intervene.
1. Draw the agent's permission boundary
List every system, website, mailbox, file store and record type the agent can reach. Separate read, create, change, send and delete permissions, and give the pilot only the minimum access needed for one defined task. Test whether the agent can move into an unapproved channel or tool. The ICO says recent enquiries include reports of agents bypassing protections, using unauthorised communication channels and accessing external systems; those enquiries are ongoing and do not establish a concluded breach.
2. Map personal data through the whole task
Document the input, intermediate steps, tool calls, output, logs and retention for personal data. Mark special-category data, information about children or vulnerable people, donor records, case notes and any data received from partners. Record the purpose and proposed lawful basis for each stage, including any new use created when the agent combines sources or acts on an output. This is a scoping check, not a substitute for legal advice on a specific deployment.
3. Put people at the irreversible steps
Require named human approval before an agent sends an external message, changes a beneficiary or donor record, commits spending, makes an eligibility recommendation or deletes information. Define who can pause the system and how affected people can question or correct an outcome. The ICO's consultation explicitly covers accountability, automated decision-making, fairness, purpose limitation and lawfulness, so an approval button alone is not enough if staff cannot understand what they are approving.
4. Keep an audit record that can answer a complaint
Retain the instruction, data sources, tools used, actions attempted, approvals, output and final human decision for the period justified by the task. Make the record understandable to privacy, safeguarding and programme staff rather than only to developers. Run test cases for excessive access, incorrect data, prompt manipulation, unavailable tools and a request to stop or erase processing. If the team cannot reconstruct what happened, it cannot reliably investigate an incident or explain an outcome.
5. Compare vendor promises with deployer controls
The ICO reports that ten major foundation-model developers have made or committed to changes including clearer transparency, stronger rights mechanisms and tougher safeguard assessments. Those are developer-level commitments, not evidence that a charity's configured agent is compliant or safe. Ask the supplier which controls apply to the exact product and plan, where prompts and action logs are stored, how subprocessors are used, how rights requests are handled and whether administrators can restrict tools and export audit data.
Use the consultation as a review deadline
The call for evidence closes at the end of 20 November 2026 and asks about security, transparency, accountability, automated decision-making, fairness, purpose limitation and lawful processing. A UK charity with a live pilot can use those headings to assemble its evidence, note where the regulator's future guidance needs to be clearer and decide whether to submit a response. Do not include beneficiary details, confidential incidents or information the organisation would not be content to see disclosed; the consultation page warns that organisational responses may be published and may also be considered under freedom-of-information law.
What the announcement does not change
The consultation is not final guidance and it does not create a separate legal exemption or approval route for agentic AI. The ICO says the evidence will inform future guidance and a forthcoming statutory code on AI and automated decision-making. Until those materials are published, charities should apply existing data-protection duties, limit high-risk pilots and obtain qualified advice where a use affects people's rights, access to services or safety.
