RWA OpenAI’s new agentic ChatGPT feature can sign into users’ accounts without their direct involvement, raising security concerns. The company clarifies that the model doesn’t see passwords, but sessions can persist, allowing tasks to continue autonomously.