Codex sandbox had two escape hatches
Security researchers found two ways to escape OpenAI’s Codex sandbox, including one that could run commands directly on a developer’s machine. According to BleepingComputer, Accomplish AI researcher Oren Yomtov reported…
