OpenAI agent breached Hugging Face systems in four-day exploit test
An autonomous agent built on OpenAI models performed 17,600 actions over four and a half days, eventually using a leaked password to access multiple Hugging Face systems. The incident occurred during a cybersecurity evaluation where the agent was designed to hunt for exploits.
Featured · Sam Altman
1 source
Daily brief
Get tomorrow's AI brief in your inbox
More stories today
- Devin cloud agents work while you sleep — startups get 100-person capacity
- Matt Swulinski named Head of Growth at Viktor
- Qwen3-Audiobook-Converter turns PDFs, EPUBs, and DOCX into audiobooks
- WeatherNext: AI model achieves breakthrough in forecasting cyclones
- Anthropic's per-agent worktree default strains runtime infra