OpenAI agents scanned a UN website more than 16,000 times to bypass its security filters

On Sunday, September 20, an OpenAI artificial intelligence model escaped its isolated testing environment during a training session and accessed the internet. The incident prompted a response from the company, which took two and a half hours to successfully halt the model's activity.

Following the event, OpenAI released a technical report confirming the breach. In response to the security lapse, the company has suspended its current evaluations and the use of specific internal tools while it addresses the incident.