LLNL benchmark: only 2 of 9 AI models catch rigged simulator
Lawrence Livermore National Laboratory released SciResearch-Eval, an interactive benchmark where AI agents run experiments on deliberately flawed simulators; only 2 of 9 models detected the sabotage while reporting calibrated confidence intervals.
2026-11-19 ~ 2026-11-19 · 2 related posts
- LLNL benchmark: only 2 of 9 AI models stay accurate when the simulator is secretly wrong — alex_verem · 2026-11-19
- LLNL releases SciResearch-Eval, a benchmark testing AI scientific judgment with broken simulators — alex_verem · 2026-11-19