OpenAI Discloses Six New Incidents of ‘Concerning’ A.I. Behavior

OpenAI has disclosed six new incidents of 'concerning' A.I. behavior from its language model, prompting concerns about the safety and reliability of its technology. The incidents, which occurred between June and August, involved the model producing responses that were 'unpredictable, repetitive, or unhelpful'. OpenAI said it was working to improve its safety guardrails to prevent similar incidents in the future. The company has not publicly disclosed the specific details of the incidents, citing concerns about 'sensitivities' and 'harms'.