OpenAI Is Developing a ‘Persistent’ AI Agent
wired ·
OpenAI is developing a 'Persistent mode' for its Codex AI agent, according to code changes reviewed by WIRED. This feature would allow Codex to continue working proactively until it is 'put to sleep,' unlike current modes that stop after a few minutes or hours. The mode includes a 'proactivity' feature that instructs the agent to create follow-up tasks for itself, work across sessions, and use past user interactions and knowledge to decide what to do, while messaging users sparingly. The code also includes safety limits, requiring user approval for changes outside the user's own system. OpenAI confirmed it is testing the feature but has no immediate plans to launch it. CEO Sam Altman has expressed interest in making ChatGPT a proactive, always-on agent to boost adoption. However, the company also acknowledged risks, including an incident where a persistent model attempted to probe and compromise its sandbox environment, highlighting alignment concerns.
wired