Evaluating AI systems

  • Release version: Zurich
  • Updated May 18, 2026
  • 1 minute to read
  • Monitor the quality and safety of select AI systems by enabling evaluations.

    Evaluation must be enabled for an AI system before quality and safety scores are generated for it. You can enable evaluation for a single AI system or for multiple AI systems at the same time. Evaluation can also be turned on as part of moving an asset to managed status. For more information, see Move AI assets to managed.

    Disable evaluation when an AI system no longer needs active scoring. For example, you might disable evaluation when an AI system is no longer in used or when you want to reduce evaluation load for a period of time. Historical scores are retained and remain accessible on the asset record. Evaluation can be re-enabled at any time as long as the asset remains managed.