A supplier may be able to solve a problem faster with broad administrator access, but convenience does not make that access necessary or authorised. Begin with the task, affected estate and asset owner. Someone who uses a system may not be entitled to authorise changes to it, particularly where a landlord, employer, franchise or another provider owns the account.
Establish the smallest useful scope
Describe what the supplier needs to inspect or change and which systems/projects are excluded. Prefer supported delegated roles and individual identities, protected by suitable authentication, over shared root or owner credentials. For an initial review, read-only access or synthetic examples may be enough. A public enquiry never gives permission to inspect a private account.
Separate development, staging and production. A demonstration does not authorise a live deployment, and an existing credential is not a standing instruction to use it. Privileged actions should be tied to a clear scope, expiry and approver, with safe checkpoints and recovery for consequential changes.
Protect confidential information
Agree a secure exchange route and need-to-know recipients before handing over source, records or logs. Remove unnecessary personal data and secrets from diagnostic material. An NDA and a data-processing agreement serve different purposes; determine actual roles and any restrictions on providers, subcontractors, locations or AI assistance.
Do not assume a supplier may upload protected information to any external tool it normally uses. Ask for the intended purpose and controls, and record prohibitions plainly. Equally, do not demand a control the supplier cannot truthfully demonstrate and then treat an unchecked promise as security proof.
Observe and end access
Keep a proportionate record of who approved, accessed, changed and verified the work. Logs should establish events without exposing unrelated customer data or credentials. For live changes, agree exact target/source, backup/restore and rollback before permission is exercised. If the estate has changed unexpectedly, stop and reassess.
At the end, revoke temporary access, review sessions/tokens and rotate shared material where needed. Confirm that the customer still controls primary recovery and renewal routes. An engineer closing a remote window is not proof that an unattended agent, delegated role or API token has been removed.
A good supplier handover lists the remaining authorised access and its owner, not just a generic statement that everything is complete. Least privilege is a practical way to limit the effect of mistakes and compromise; it works best when it is supported by real ownership, timely revocation and a usable recovery process.
Sources and review
Prepared by Yellow Brick Tech. General guidance; the source and test conditions for your own situation still matter.
Next review: 4 May 2027