According to OpenAI... New guidelines detail threat models, controls, and audits for frontier RL training to reduce sabotage and data theft risks.

According to OpenAI... the company outlines safety cases for securing frontier RL training runs and risk controls for high-stakes model training.