None of these are refusals on principle. Each is a case where signing would mean asserting something untrue, and we would rather have the conversation in week one than be found out in an incident review.
A warranty that we hold a certification. We hold no security certifications — no external audit, no attestation. A clause that warrants one, or that requires us to maintain one throughout the term, is a clause we cannot sign. We will describe and evidence the controls instead, and if your process has a hard certification gate we will tell you on the first call.
A commitment to a control we do not operate. A twenty-four hour staffed security operations centre, a formal third-party screening programme, a documented control set that has been externally tested — we do not have these, and a DPA that says we do is worthless to you. Where your policy genuinely requires one, that is a real gap and worth knowing about early.
Liability for a sub-processor we do not contract. If you hold the agreement with the model provider and the cloud, their acts are not ours to indemnify. We will help you read their terms and configure their retention settings, and we will name them in the architecture document, but we cannot underwrite a contract we are not party to.
An unqualified on-site audit right. We will answer a written questionnaire in full, put the engineer who built the system on a call with your reviewer, and walk your auditor through the architecture, the logs and the access model. An unrestricted right for any of your customers to audit us at any time, at our cost, is not something a firm of our size can honestly commit to, and we say so rather than signing it and hoping it is never exercised.
A processor designation for data we never touch. Covered above, and the most common one. It is easy to sign and quietly wrong.