Framework detects hidden policy rules in commercial AI models

Original: Auditing Proprietary Alignment in Large Language Models: A Comparative Framework Without a Ground-Truth Standard

Source: arxiv.org ↗