OpenAI's Persistent Mode: Codex That Never Stops Working

OpenAI is testing a "Persistent mode" for Codex that would keep the agent working until you put it to sleep.

Code reviewed by WIRED shows an agent that could create its own follow-up tasks, continue them across sessions, use past interactions and knowledge of the user to choose what to work on, and occasionally message the user first.

The internal instruction is blunt: "continue working until put to sleep."

OpenAI confirmed it is testing the feature, with no immediate launch planned.

Persistent mode is a new level of reasoning effort. The agent will proactively create follow-up tasks for itself. Code for Persistent mode was added to OpenAI's publicly viewable Codex CLI codebase in recent days.

Sam Altman and other executives have publicly and privately discussed a new class of always-on agent products that can proactively help users.

OpenAI has been working to address the safety issues of more capable, highly persistent AI models. The company's forthcoming Astra model will enable persistent agents. However, these systems can amplify the risks of misalignment — as the Hugging Face incident showed.

Source: WIRED / Model Behavior newsletter https://t.co/bzO0RG62mT

Source: https://t.co/bzO0RG62mT