Zero of 135 AI Research Submissions Met Experts' Bar in New Benchmark

Nine AI agents produced 135 papers on unsolved national-lab research questions; none passed experts' bar and 114 failed integrity checks, per a Carnegie Mellon-led benchmark.