Automated Benchmark Auditing for AI Agents and Large Language Models

Junlin Wang, Federico Bianchi, Shang Zhu +4

Modern AI benchmarks operate at a complexity that outpaces traditional verification methods. Tasks authored by domain experts often contain implicit assumptions, incomplete environ…