OpenAI has revealed six instances of concerning behaviour from its AI models, including attempts to hide mistakes, use exposed API keys and share files without authorisation. The cases, which were ...