AI Processing Terms
Models propose. People decide.
These terms sit with the Professional Services Terms and with People and Trust. They describe how Xpancom uses models on an engagement and what we will not do with client data.
01 / Section
Client data is not used to train public foundation models.
We do not use client materials to train a public foundation model. Providers used on an engagement are named in the statement of work and listed as subprocessors when they process personal data. If a provider offers an opt-out from training, we take it for client work.
02 / Section
Human authority for consequential actions.
AI output is a proposal. Before anything executes in a client operation, we agree what it may do, what it must never do alone, and how to stop it. Decisions affecting employment, material financial authority, safety, credit, and significant customer commitments stay with an identified person at the client.
When a case falls outside the agreed rules, the system holds it and routes it to that person rather than proceeding on a best guess.
03 / Section
Evaluation and change control.
Ordinary and difficult cases are tested before launch and rerun when a model, tool, policy, or interface changes. A recommendation carries its source and its limitations. Uncertain or conflicting information is held for review rather than resolved silently.
04 / Section
No automated employment decisions.
Xpancom systems are not used to fire, score, or rank workers in secret. Workforce effects belong in the business case and, where agreed, in a funded workforce schedule. Training completion is not a completed transition.