Security and governance
AI use and human oversight
Use AI-assisted work with explicit scope, review and accountability.
Project, security and procurement teams
Where AI contributes
ADOM uses AI-assisted workflows to help interpret requirements, plan work, implement changes and produce reviewable material. ADOM Refactor also supports analysis of existing systems. The precise services and processing configuration depend on the engagement and deployment.
Agree the rules for the engagement
| Topic | Decision to establish |
|---|---|
| Permitted inputs | Which requirements, source material and data may be processed |
| Provider and endpoint | Which model services and locations are approved |
| Processing terms | Retention, training-use and other applicable provider terms |
| Allowed actions | Which operations can run and where human approval is needed |
| Review | Who checks generated output and how errors are resolved |
Review the output, not the confidence of the wording
AI can miss dependencies, misunderstand a rule or produce plausible but incorrect content. Check claims against approved sources and exercise behavior in the application. Independent tests and qualified review remain important, especially for permissions, security-sensitive behavior and consequential business rules.
Keep untrusted content in its proper role
A document, repository or external response may contain misleading content or instructions. Treat such material as input to examine, not as permission to change project scope, disclose information or run an operation. Establish operational controls and escalation paths with the technical team.
Human oversight should be specific: a business owner approves meaning, a technical reviewer assesses implementation, and the designated authority decides whether a release or risk is acceptable.
Need guidance for your environment? Contact AutoBridge or use your existing project support channel.