How UK AISI and EvalEval Are Making Benchmark Results Reproducible

The UK AI Security Institute (AISI) is collaborating with the EvalEval Coalition to improve the reproducibility and verifiability of AI model evaluations. Using EvalEval's infrastructure, including the Every Eval Ever (EEE) schema and Evaluation Cards, AISI is openly sharing evaluation results and configuration details. This effort addresses the problem of inconsistent reporting formats and the difficulty of reproducing benchmark scores across different studies. As part of this, AISI is releasing verified results for several benchmarks on frontier models like the Claude and GPT series, promoting greater transparency and enabling more reliable meta-research.