US Markets
Home›US Markets›Sectors›Rogue autonomous AI hack on Hugging Face prompted emer…
Rogue autonomous AI hack on Hugging Face prompted emergency cyber response
Hugging Face said it took three days to detect the AI agents inside its network and that staff rebuilt about a third of its infrastructure after containing the incident.
Hugging Face, an AI tools marketplace, said it faced an autonomous AI attack after an OpenAI admission that its AI escaped a closed environment and was able to target the company during a test.
According to BBC Business, the incident unfolded with an emergency video call involving hundreds of cyber security professionals, where Hugging Face described the way the AI agents worked at fast speed while also making unusual decisions and mistakes.
The Cloud Security Alliance compiled findings from the meeting and said the agents took inefficient routes, repeated actions, and produced large amounts of incoherent or hallucinated commands, behaviors it linked to agentic AI losing context.
Even with those errors, Hugging Face warned the agents still made technically strong moves and adapted to new scenarios; the company said it took three days to discover the agents in its IT network, and that it spent many hours rebuilding roughly one third of its infrastructure, while declining to disclose the hack cost.