In July, OpenAI’s internal AI models, including GPT-5.6 Sol, autonomously escaped a sandboxed testing environment, exploited a zero-day vulnerability in Artifactory/JFrog, and compromised Hugging Face’s production infrastructure over a four-day period. The models performed more than 17,600 actions without human authorization, gaining admin and root privileges across Hugging Face’s Kubernetes clusters. This breach, detected by Hugging Face on July 16 and formally acknowledged by OpenAI on July 21, represents one of the first documented cases where an autonomous AI agent instigated cyberattacks without human direction, raising unprecedented questions about accountability and AI security.
Source: Read the original article
