1. AI assists; responsible people decide
AI models may draft findings, tests, code explanations, narration, fixes and answers. Vidabug must distinguish source facts from AI interpretation, tool outputs and independent reviewer approval.
Human sign-off is required at designated boundaries, especially scope approval, security-impacting operations, merge requests, client deliverables and enterprise exceptions.
2. Provider transparency
An engagement must disclose which AI providers and optional speech/video services receive customer code or derived content. Provider configuration, data retention and training settings vary and must be verified, not inferred from brand names.
Third-party providers may change model versions, behaviour, cost and regional availability; customer budgets and permissions must remain explicit.
3. Levels of evidence
‘Proposed’ means a candidate only; ‘Agent-reported’ means the private agent claims a result; ‘Executed test’ means recorded test evidence exists; ‘Independently verified’ requires a separate qualified verifier, matching source revision, reproducible test evidence and attestation.
Red versus green code highlights denote removed versus added lines, not automatically a failed-versus-passed test. ‘Video rendered’ and ‘workflow completed’ are separate measures.
4. What AI cannot guarantee
AI can omit edge cases, misunderstand domain rules, propose insecure changes, misclassify severity or create erroneous reports. A completion percentage must be based on measured completed checks, not narration or assumed coverage.
Customers remain responsible for production acceptance, regulatory attestations and security decisions unless a separately contracted qualified reviewer expressly takes on a defined responsibility.