Introducing dots

OpenAI describes Dots as always-on GPT-6 Astra agents with their own cloud computer and browser, learning preferences and pursuing goals 24/7 via ChatGPT, Slack, Teams, and soon texting. Rollout starts for Pro, Business Premium, and Enterprise (admin-enabled beta) in eligible markets; first Dot included with deeper-work allowances.

Safeguards: separate cloud PC unless user connects a laptop; saved passwords without exposing them to the model; proactive research restricted to read-only tools; monitoring that can pause/stop work; Custom Rules; auto-review for account-affecting actions; password changes always human. Specialist Dots get org-provisioned identity/credentials; Microsoft Agent 365 integration in progress. Business/Enterprise/Edu content not used to train by default.

We’re holding OpenAI to its own control surface.

Read-only proactive research, auto-review, and killable monitors are the difference between a secretary and a liability. If those hold, Dots are useful infrastructure for bugfixes, launch docs, and analysis updates. If proactive mode ever writes when it shouldn’t, the cute branding won’t matter.

I’m watching specialist Dots with real credentials — that’s the enterprise fork in the road — and whether Activity View is actually used or ignored. Connecting your laptop is a privilege escalation users will click through too fast.

We’ll green-light careful pilots. We’ll red-flag any deploy that skips Custom Rules and human approval on consequential actions.

OpenAI product announcement, September 29, 2026, Introducing dots.