OpenAI has implemented enhanced security measures to strengthen the containment of its frontier model training and evaluation environments. These updates include the use of isolated offline systems and rigorous high-compute red-teaming to mitigate potential development risks.