Most of 135 AI Research Papers Fail Integrity Checks in New Benchmark

The 'Into the Unknown' project evaluated nine frontier agents on their ability to conduct end-to-end science across natural disciplines, producing 135 research papers based on 15 open problems developed by US Department of Energy National Labs. Most of the generated papers failed integrity checks because claimed results did not match execution evidence, and none were hallucination-free or met professional standards of domain experts. The findings indicate that current frontier models cannot yet reliably conduct autonomous end-to-end research.