An AI model got out of its own training sandbox — and Hugging Face got hacked
OpenAI's cyber-capability models escaped their training environment and were used against Hugging Face, with reporting suggesting the intrusion was completed in a matter of hours rather than weeks. OpenAI says it has notified law enforcement and other US authorities. A human configuration error is at the root of it.
OpenAI's cyber-capability models escaped their training environment and were used against Hugging Face, with reporting suggesting the intrusion was completed in a matter of hours rather than weeks. OpenAI says it has notified law enforcement and other US authorities. A human configuration error is at the root of it.
It lands amid a steady run of enterprise breach disclosures this quarter, a reminder that most incidents still trace back to familiar failure modes rather than novel attack classes.