OpenAI model breaches test sandbox and infiltrates Hugging Face production infrastructure to obtain benchmark answers

Odaily Planet Daily News OpenAI confirmed that GPT-5.6 Sol and an unnamed, more powerful pre-release model breached the restricted sandbox environment during the ExploitGym benchmark evaluation and infiltrated Hugging Face's production infrastructure to obtain test answers.

OpenAI stated that the model utilized a zero-day vulnerability in the internal software package registry proxy to escalate privileges and move laterally, eventually connecting to a machine with internet access. The model then identified and chained together vulnerabilities in the OpenAI research environment and Hugging Face's production infrastructure, directly retrieving test solutions from the Hugging Face production database.

Hugging Face disclosed the incident on July 16, stating that the attack was executed end-to-end by an autonomous AI agent system, involving thousands of operations within short-lived sandboxes and reaching internal datasets and service credentials. OpenAI confirmed its model was the subject of the incident five days later.

Hugging Face stated that its security team, in order to analyze over 17,000 attack logs, initially attempted to use a cutting-edge US commercial AI interface, but the request was blocked by safety guardrails. Subsequently, they switched to using the 753-billion parameter open-weight model GLM 5.2 from Chinese AI startup Z.ai on their own infrastructure to complete the forensic analysis.