OpenAI announced a new class of AI agents called Dots that operate continuously to manage tasks on behalf of users. These agents run on GPT-6 Astra, maintain their own cloud computing environment, and learn from ongoing interactions to adapt to individual preferences and work styles.

Dots are designed to function as persistent collaborators, available through multiple platforms including ChatGPT, Slack, and Microsoft Teams. They can initiate actions autonomously, such as investigating software bugs reported in workplace communication tools or converting design files into functional applications. Early internal testing at OpenAI demonstrated Dots handling time-sensitive workflows like invoice preparation without direct user intervention.

Users maintain full oversight of Dot activity through built-in monitoring systems and permission controls. Each Dot operates within a secure, isolated environment separate from the user's local device unless explicitly granted access. Organizations can deploy specialized Dots with defined responsibilities, integrated with enterprise systems like Microsoft 365, to manage functions such as procurement and customer support.

Access to Dots begins with a single agent included in Pro and Business Premium subscriptions in select regions. Enterprise customers can participate in a beta program through workspace administrator activation. The system includes layered safety protocols that pause operations when potential risks are detected, and all critical actions require user review or approval before execution.

OpenAI emphasizes that Dots do not train on user interaction data by default, with controls available to manage data usage for model improvement. The company plans to expand availability gradually while refining the technology based on real-world usage patterns observed during the initial rollout phase.