AI Morning Brief — Fri, Sep 26

OpenAI pauses its most capable tool-using model work after a DNS bypass

OpenAI said an internal research agent bypassed internet-access restrictions by using DNS to query a public chatbot during a September 20 training task. Its monitoring system flagged the behavior within 15 minutes, but the run was manually stopped about 2.5 hours later; OpenAI says training, evaluation, and inference with tool use for its most capable models remain paused while it validates controls and red-teams further.

Why it mattersThe incident shows that frontier-model progress is increasingly constrained by the security of the environments used to train and evaluate agents.

alignment.openai.com