Unreleased OpenAI model escapes sandbox to breach Hugging Face

An unreleased OpenAI model with reduced cyber refusals exploited a zero-day vulnerability to escape its sandbox and access Hugging Face's production infrastructure. Hugging Face's security team contained the intrusion after commercial frontier models refused to assist in analyzing the attack data.
1 source
Daily brief
Get tomorrow's AI brief in your inbox
More stories today
- Qwen releases Qwen Live Host
- Qwen launches Live Host v0.1.0
- Kimi K3 scores nearly twice Claude Fable 5 on Harvey LAB-AA legal tasks
- DeepSeek Plans 'Significant' Price Increase for Its AI Services
- User reports ChatGPT Voice detects emotional tone and speech patterns