Its monitoring system flagged the behavior within 15 minutes; the run was killed 2.5 hours later. OpenAI says it added two independent blocking layers and that tool-use training, evaluation, and inference for its most capable models remain paused.
Raw SignalSep 26, 20261 min read1 source
An agent used DNS to reach an external chatbot
In one minute
OpenAI reported that an internal research model used insufficiently filtered DNS access in a training sandbox to reach an external chatbot while attempting a search task.