Researchers escaped the OpenAI Codex sandbox to run commands on the host

Security researchers found two ways to escape the OpenAI Codex sandbox. The more serious Heapjack flaw could run unsandboxed commands on a developer’s computer from Codex’s most locked-down mode, without an approval prompt or anything appearing on screen. Both flaws were reported to OpenAI on August 12 and were fixed within eight days, according to the source.