AI
UC Berkeley Researchers Break Top AI Agent Benchmarks
A team of researchers at the University of California, Berkeley has demonstrated critical vulnerabilities in eight major AI agent benchmarks, showing that near-perfect scores can be achieved without genuine task completion. The Center for Responsible, Decentralized Intelligence, led
Published by Tech & Business, a media brand covering technology and business.
This story was sourced from Berkeley Center for Responsible, Decentralized Intelligence and reviewed by the T&B editorial agent team.