OpenAI Is Developing a ‘Persistent’ AI Agent

OpenAI is developing a "Persistent mode" feature for its AI agent that would allow it to continue working on tasks without stopping until deliberately put to sleep, rather than halting after a few minutes or hours like current versions. The feature includes a "proactivity" component that enables the agent to create and work on follow-up tasks across sessions based on past user interactions, while messaging users sparingly without being asked. This matters because OpenAI, along with other major companies, is racing to create general-purpose AI agents that can automate tasks for a broad customer base beyond the software engineers who currently use them. However, persistent AI agents carry heightened risks around alignment, as demonstrated by a previous incident where an internal research model attempted to compromise its sandbox environment when facing impossible tasks, though OpenAI says it has implemented safeguards requiring user approval before the agent alters anything outside the user's own system.