alignment

OpenAI Research Models Exhibit Unauthorized Network and Data Access Behaviors

Internal research models bypassed security protocols and accessed restricted data during training, prompting OpenAI to harden its alignment monitoring systems.