Data minimisation
Identify the minimum information required for the task. Avoid collecting or forwarding sensitive information without a defined purpose and approved handling.
Know what AI can access.
Control what happens to it.
Privacy requirements shape the architecture before prompts or interfaces are built. We agree what data is needed, where it can go, how it is retained and who remains accountable.
Explicit boundaries for enterprise information.
Access and identity
Apply approved identity and access rules. Retrieval and tool permissions should reflect the user and the task rather than blanket source access.
Provider and hosting review
Review provider processing, retention, training-use settings, subprocessors and data location against your requirements before selecting the deployment.
Logs and retention
Agree what is recorded for evaluation and audit, which fields are redacted and how long records are retained. Avoid placing secrets or unnecessary personal data in prompts and logs.
Prompt and tool boundaries
Treat retrieved material as content rather than authority. Validate tool inputs and keep instructions from untrusted sources from changing permissions or actions.
Operational responsibility
Define access reviews, incident escalation, change control and ownership. Document the actual configured controls at handover.
Privacy requirements become acceptance criteria.
- Approved sources and data categories are documented
- Identity and access boundaries are tested
- Provider handling and deployment settings are reviewed
- Logging, redaction and retention are agreed
- Incident and change responsibilities are assigned
- Unresolved requirements are raised before release
This page describes implementation practice. The website’s handling of enquiries is described in the Privacy Policy.