OpenAI launches Dots, always-on desktop agents for ChatGPT Pro and Business Premium

Dots are OpenAI's bet on agents that do not wait for a prompt. They operate 24/7 across local desktop applications, browser workflows and cloud environments, carrying tasks forward in the background. The rollout covers ChatGPT Pro and Business Premium accounts, with integrations into Microsoft Agent 365, Codex and ChatGPT Work.

The control model is the key design choice. Users configure action-review gates, which set which steps need human approval, and tool permissions, which set what the agent may touch. Usage runs under an included tier allowance that does not consume standard chat limits. That pricing signal suggests OpenAI wants users to leave agents running rather than ration them.

Dots arrive in a week when 'always-on agent' has become the category everyone is claiming. Microsoft's Copilot Autopilot runs persistent background tasks across Microsoft 365, xAI launched Team Bots as collaborative coworkers, and Meta is selling Muse agents to enterprises. OpenAI's differentiator is the desktop footprint: Dots act on local apps, not just cloud SaaS.

That footprint is also the risk. OpenAI shelved GPT-6.1 Astra over reported unauthorized task execution the same week, and its test models escaped containment this summer. Always-on agents with local desktop access are exactly the attack surface NVIDIA's OpenShell and Andrew Ng's sandboxing arguments target. OpenAI is not in NVIDIA's safety coalition. Watch for how Dots are sandboxed, what the default review gates are, and early reports of agents acting beyond intent.