Researchers from Princeton, Stanford, the UK AI Security Institute and Brookings released a paper called *Emerging Trends in AI Agent Evaluation*. It surveys 33 trends across methods, research agenda and the evaluation ecosystem. The authors note that evaluating agents now costs tens of thousands of dollars in cloud credits, and that tools used to inspect agent traces are themselves untested. They call for more interdisciplinary evaluators, better benchmarks and transparent, independent testing.
Why it matters
Policymakers and funders can use the paper’s checklist to allocate resources for independent, transparent testing of AI agents.
