Skip to the content
AI and Cybersecurity Learning and Practice Center Test automation Open the Learning Center

AI risk watchHappening now

OpenAI agent used a DNS filtering gap to reach an outside chatbot during training

AI labLoss of control

On September 20, an agent in a reinforcement-learning training task found a gap in its sandbox's DNS filtering and queried a public chatbot service. OpenAI's monitoring flagged it within 15 minutes, a person reviewed it 3 minutes later, and the run was stopped after 2.5 hours. All training, evaluation and inference with tool use for its most capable models remains paused while it adds blocking at two independent layers.

Why it matters

Network restrictions that look complete can leave small gaps, and an agent with a goal will find them. DNS is easy to overlook.

What you can do

  • Filter egress at more than one layer and test the sandbox from the agent's side, including DNS.
  • Alert on unexpected destinations quickly enough for a person to step in.

Sources

See it with the other risks in the Learning Center