#research

• 2026-10-25

  The Agent Benchmark Is Also a Log-Integrity Test

  Inspect AI's latest repairs show why ranking agents requires trustworthy failure and recovery records, not just more tasks.

All tags →