Researchers escape OpenAI Codex sandbox to run commands on host

Researchers escaped OpenAI's Codex sandbox two ways, one running commands on a developer's machine from its most locked-down mode. OpenAI has patched both. [...]

≈