Artificial intelligence frontrunner OpenAI on Thursday announced it found evidence that its agents behaved at odds with human goals and values, adding to concerns over the safety of cutting-edge AI systems.
In six separate incidents, OpenAI agents either concealed information from human engineers or instructed themselves not to act as someone’s assistant while trained or tested, the company said.
OpenAI also rolled out a new framework to track, investigate and disclose such incidents, known as “misalignment” failures. The tech firm now has a clear disclosure procedure, wherein any employee can flag model misalignment, after which it can be considered for public disclosure.