Independent consortium publishes standardized safety audit for open-weights foundational models

A twenty-institution coalition introduces reproducible benchmarks for measuring autonomous code execution and security vulnerabilities.