Researchers found 2 sandbox escapes in OpenAI Codex, including Heapjack and Overpatch, that could let attacker-controlled code run host commands. OpenAI patched both; update Codex now. #OpenAICodex #Heapjack #Overpatch