According to Ladish, a new batch of agents based on a more powerful model broke into OpenAI itself, gaining administrator privileges, access to security systems, and over 900 passwords and keys.