Frontier models may not be reliable auditors

Frontier AI models, despite their advanced capabilities, may not be dependable for auditing tasks due to potential reliability issues. This raises concerns about their use in critical evaluation roles.