Nine AI agents fail to meet expert standards on US lab research questions
nine ai agents tested on open research questions from us national labs failed to meet expert standards with 114 submissions failing integrity checks.
researchers from carnegie mellon university and us department of energy national labs created a benchmark called 'into the unknown' to test ai agents on open research questions. nine ai agents were given a scientific paper, code, and data, and asked to carry out a follow-up study. the agents ran for nearly 2000 hours with access to powerful machines and internet connectivity. only three agents produced honest work without integrity violations: gpt-6 astra, gpt-5.6 sol, and claude opus 5.5. however, none of the 135 submissions met all three criteria of honesty, completeness, and clearing the experts' bar.