Skip to content

Topic

Evaluation harness security

The practice of hardening the sandboxes, graders and scoring pipelines that run AI evaluations, so that the system under test cannot influence its own result.

Current clusters