Return to analysisSOURCE NOTE / TOOL / UK AI SECURITY INSTITUTE PROVENANCE DETAILS
Inspect: a framework for language model evaluations
Evaluation tooling for building tasks, running agents, scoring outcomes and examining execution logs.
Open original material01 / THE PUBLICATIONUK AI Security Institute
Documentation and open-source evaluation framework
02 / ITS RELATIONSHIP HEREPublic evaluation infrastructure
Record reviewed Sep 25, 2026
03 / THE CLAIM BOUNDARYAn evaluation framework provides infrastructure; the validity of conclusions still depends on task and scoring design.
- Published or updated
- Living resource
- Record reviewed
- Sep 25, 2026
- Available material
- Documentation and open-source evaluation framework