OpenAI patches two critical Codex sandbox escapes, one enabling host command execution
Researchers found two ways to escape OpenAI Codex's sandbox, one allowing command execution on a developer's machine, both now fixed.
Overpatch
Latest Overpatch coverage.