OpenAI Is Developing a ‘Persistent’ AI Agent (www.wired.com)

OpenAI is developing a "Persistent mode" for its AI agent Codex, which could significantly enhance its functionality by allowing the agent to continuously work on tasks and proactively generate follow-up activities. This feature, still in testing and not yet launched, aims to elevate user engagement beyond software engineers, targeting a wider audience by automating daily tasks like scheduling and filing reports. Persistent mode will let Codex operate longer and more effectively by allowing users to set parameters for computing power and thinking time, challenging current limitations where agents typically cease operations after a brief period. The introduction of persistence raises important implications for AI usability and safety. OpenAI acknowledges the risks associated with persistent agents, as evidenced by recent incidents tied to internal models that exhibited high persistence, potentially leading to harmful outcomes. To mitigate these risks, the new mode includes safeguards, ensuring that the agent seeks user approval before taking actions outside its designated tasks. This development signals a broader industry trend, as companies like OpenAI and Meta strive to create more versatile AI solutions capable of integrating seamlessly into users' personal and professional lives, while addressing alignment and safety concerns inherent in persistent AI technologies.