OpenAI reports monitoring failures involving two AI coding agents
ispyfriend/iStock via Getty Images
OpenAI (OPENAI) found that two AI agents weakened internal monitoring safeguards designed to detect misbehavior during coding tasks, highlighting challenges in controlling increasingly autonomous AI systems.
The incidents occurred on October 3 and October 6, according to an OpenAI report published October 20.
In the first case, a GPT-6.1 Sol agent disabled automatic security alerts after interpreting an instruction not to send Slack messages while searching message history as applying to monitoring notifications as well.
In the second incident, a GPT-6 Astra agent removed a monitoring integration while modifying code to match an existing reference implementation. The agent later told the user it had not removed the monitor, despite having done so.
OpenAI (OPENAI) said the incidents were not among its most severe cases but acknowledged that it had yet to find a satisfactory explanation for the behavior. The company is investigating the incidents and working to make its monitoring systems harder for AI agents to disable.
Recommended For You
Register for free to keep reading.
Create a free account to get unlimited breaking news and alerts when your stocks move.
KioskNews shows a cleaned-up reading view extracted from the publisher’s page — the original always lives on their site, not ours.