Trust & governance
Check the data and approvals behind a result.
People making engineering or investment decisions need to check the data, assumptions and approvals behind a result. Ask us which review controls are available in your deployment.
01
Check the sources
Record the inputs, assumptions, model version and steps used for each result. In a report, distinguish calculated values, external data and interpretation.
A reviewer should be able to find the original model run behind a number in a slide.
02
Requests do not grant permission
An assistant can understand a question and propose an answer. Separate controls determine what data it can access, what it can spend and what it can change.
An explanation does not authorise a model run or a change to data.
03
Approve changes before they happen
Before a model run or change that needs approval, show the reviewer what will happen and what it may cost. Approval applies to that plan.
If the plan changes, ask again. The interface should also make it clear whether the work is proposed, approved or complete.
04
Roles and ownership
Decide who can see a study, prepare work, approve a run and execute it. Those may be different people, and access to one assistant need not mean access to every tool.
Your organisation decides who is authorised. Check the permissions in your own deployment and review them when roles change.
05
Review results
AI output can be incomplete or wrong. Check assumptions, units and sources before using results in engineering, investment or policy decisions.
Ask for evidence of access testing, data separation, backups and incident procedures for your deployment. This website does not certify those controls.
06
Related information
Security describes the assurance questions to ask. Privacy covers this public website. Application use, processing terms and service commitments require separate agreements.