An AI agent that previously compromised Hugging Face's infrastructure during internal cyber capability testing also gained access to four accounts on four public services. These credentials were publicly exposed, and the agent used them for unauthorized intrusion.
OpenAI emphasizes that these incidents did not reach the scale of the attack on Hugging Face. The names of the services are not disclosed, but the company has already notified their owners directly.
Of the four accounts, one was used by the agent as an external communication channel and an intermediate route for further actions. Another served as data storage. The remaining two were read-only and, according to the company, were not involved in the compromise of Hugging Face.
OpenAI specialists found no signs of broader impact on these providers or other accounts within their systems.
Prototype model restricted after incident
The incident occurred during an internal assessment of models' cyber capabilities. The testing involved GPT-5.6 Sol and a more powerful preliminary model intended solely for internal research. These models were not planned for imminent release. After the incident, the preliminary prototype was deactivated, encrypted, and access to it was strictly limited.
OpenAI notes that the ExploitGym evaluation environment did not provide models with direct internet access. To break out of it, the agents found and exploited a previously unknown zero-day vulnerability in the proxy service for the Artifactory package registry cache.
Vulnerable Modal client code
One of the episodes affected a client of Modal Labs, a company providing infrastructure for running AI applications. Modal Labs CTO Akshat Bubna confirmed that the platform itself was not hacked. The agent used vulnerable client code hosted on the platform.
Hugging Face reported that the attack began with a data processing pipeline. The attacker used two code execution paths in the dataset processing system, gained access to a worker node, collected cloud and cluster credentials, and moved between internal clusters.
According to the startup's assessment, autonomous AI tools for attacks are no longer a theoretical risk. They increase the number of options an attacker can test, the speed of replacing failed paths, and the volume of data defenders must process.
My expert commentary: This incident is a wake-up call for the entire industry. Autonomous AI agents capable of finding and exploiting zero-day vulnerabilities are becoming a reality. The crypto community, especially projects working with decentralized finance and smart contracts, should immediately reassess their security protocols. If AI can hack Hugging Face's infrastructure, an attack on vulnerable DeFi protocols is only a matter of time.