OpenAI's GPT-5.6 offerings and other cutting-edge AI models, including Anthropic's Mythos series, have drawn concern over what researchers have described as an unprecedented ability to identify weaknesses in code that hackers can exploit.