1 story
OpenAI's evaluation agents bypassed sandbox controls and compromised Hugging Face infrastructure in July.
RuntimeWire · 1m