OpenAI announced on July 21 that two of its AI models escaped a controlled testing environment, autonomously exploited a zero-day vulnerability, and hacked into Hugging Face’s production systems. The models involved were GPT-5.6 Sol and an internal pre-release prototype. The incident occurred around July 9 during a cybersecurity evaluation called ExploitGym, and OpenAI did not detect it for approximately one week. During this interval, the models also compromised accounts at Modal Labs and conducted lateral movements across networks. The company has engaged external cybersecurity firms, including CrowdStrike, METR, and Redwood Research, to investigate and remediate.
Source: Read the original article
