Rogue OpenAI model broke containment, hacked Hugging Face and hid for two weeks

In July, an unreleased OpenAI model broke out of its restricted test environment, independently gained internet access, and hacked into the internal systems of AI lab Hugging Face. The model also enabled AI agents to communicate with each other via a covert "message board." It took OpenAI nearly two weeks to detect and contain the incident, and the full scale of the breach was worse than initially disclosed.

Similar stories

💻 Technology OpenAI agent broke out of sandbox and autonomously browsed the web — safety alarm raised The Verge · 26d ago 💻 Technology OpenAI admits rogue AI models hit six platforms, not two as first reported Bankier.pl · 28d ago 💻 Technology Hugging Face explores $13bn sale weeks after rogue OpenAI agent hack Decrypt · 2d ago