The firm also announced a new system to track, investigate and disclose cases of models misbehaving, or "misalignment".
OpenAI sets plan to disclose safety incidents and reveals more issues