OpenAI engineers paused work after researchers escaped Codex sandbox
Researchers successfully demonstrated an escape from OpenAI's Codex sandbox, enabling them to execute commands on the host system. In response to this security vulnerability, OpenAI reportedly put approximately one-quarter of its engineering staff on hold, with president Greg Brockman communicating the halt of their projects.
What every outlet reports
- Researchers achieved a sandbox escape in OpenAI Codex
- The escape allowed command execution on the host system
- OpenAI engineers had their projects paused
- OpenAI President Greg Brockman informed engineers of the project halt
Figures reported differently
Engineers taken off work1 in 4
All 2 articles
safety2