Back to standardsAI Evaluation & Release Standard
Testing requirements before and after production release of model, agent, or tool changes.
Minimum controls
- Golden test set defined
- Safety, grounding, and tool-use evaluations required
- Prompt injection and jailbreak testing required
- Regression and cost/performance testing required
- Release gate approval and periodic re-evaluation required
Related records and procedures
Records: Evaluation, Golden Test Set, Red-Team Result, Release Approval, Drift Monitoring.
Procedures: Run pre-prod evaluation, red-team review, approve release, periodic re-evaluation.