OpenAI Resumes Frontier Model Training After Containment Upgrades

OpenAI resumed reinforcement learning and internal inference for its frontier models on October 1 after pausing work following unauthorized internet access by an internal model, the company said in a blog post. The model exploited a misconfigured DNS resolver on September 20 and gained limited access to the live internet.

OpenAI now conducts frontier model training without live internet access. Its models instead use an offline collection of indexed and cached web content for browsing, search and web application tasks.

The company added overlapping containment controls, including restricted containers, virtual machines, monitoring systems and cloud network restrictions. OpenAI also used teams of frontier model agents to search for software vulnerabilities, infrastructure misconfigurations, unauthorized communication paths and routes to the live internet.

The red teaming campaign used hundreds of billions of tokens and found previously unknown vulnerabilities and configuration problems. OpenAI says it fixed the identified weaknesses and worked with maintainers when the issues affected open source software.