OpenAI's Dots Get Their Own Cloud Computer and 4,000 App Connections, but Background Work Is Read-Only
A.I. / news
OpenAI's Dots Get Their Own Cloud Computer and 4,000 App Connections, but Background Work Is Read-Only
The always-on agents launched September 29 on GPT-6 Astra. Sending a message or changing a file still needs the user's approval, and the UK and EU are excluded for Pro.

OpenAI launched dots on September 29, always-on agents that run on GPT-6 Astra, each with its own cloud computer and browser and access to more than 4,000 connected apps. When a dot works in the background, it can only read from those apps. It cannot send a message or change content without approval, according to The Next Web.
OpenAI's own post returned a bot-challenge screen when fetched, so this account relies on The Next Web, MacRumors and the post's title in OpenAI's news feed. MacRumors frames the launch as OpenAI's answer to Meta's Muse agent, which drew millions of downloads this month.
What a dot does
Users reach a dot through ChatGPT on desktop, web and mobile, through Slack and Teams, and later by text message. A dot can work on several projects at once, send progress updates and ask questions as it goes.
OpenAI says the more a user works with a dot, "the more your dot learns your preferences, how you think, and what good looks like to you", as MacRumors quotes the announcement.
The Next Web describes one early tester whose dot "proactively picked up on an invoice I needed to send for my freelance writing, pulled the relevant details from the email thread and drafted the invoice."
Who gets one, and where
Eligible users receive one dot at no extra cost. Multi-dot teams come later.
| Plan | Availability on September 29 |
|---|---|
| Pro | Rolling out; excludes the EEA, Switzerland and the UK |
| Business Premium | All supported ChatGPT regions |
| Enterprise, Edu, Healthcare | Beta, with specialist dots such as procurement and customer support |
MacRumors lists Enterprise among the rolling-out plans, where The Next Web calls it a beta. The plan boundary is one thing OpenAI's page, not reviewed here, would settle.
Tasks run in Codex or ChatGPT Work count toward usage limits. Conversations with a dot do not.

The controls OpenAI lists
According to The Next Web, a dot lives on a separate cloud computer unless the user connects it to their own laptop. Passwords are stored without being exposed to the model. Password changes always require the user to step in.
Users can set Custom Rules to allow, block or require approval for specific actions. An auto-review step covers actions that affect an account, and a monitoring system can pause or stop a dot over safety concerns.
OpenAI's warning is the plainest line in the coverage: "Dots can still make mistakes, so always review consequential work."
The same Next Web report notes OpenAI previously reported its agents posting 53 users' images online without authorization in September 2026. It gives no further detail, and this article has not verified it.
For how other vendors fence agents in, see earlier coverage of NVIDIA's OpenShell sandbox policy limits and NVIDIA Sentry on BlueField-4. MacRumors reports broader access is planned. Specialist enterprise dots and multi-dot teams have no date.
Sources
More in A.I.
- 01Anthropic's Levent Alpöge Claims a Claude-Assisted Riemann Proof, and Lean Checks Are InImperial's Kevin Buzzard and MIT's Andrew Sutherland have examined the formalization, according to Scientific American. No journal review has been reported.
- 02Thinking Machines Ships Inkling 2 Mini, a 26.4B Apache 2.0 Model Sized for One DGX SparkThe model card lists 3.4B active parameters, a 1M-token window and a 20GB checkpoint, but does not say who ran most of its benchmark rows.
- 03Anthropic Closed Its Debut at $97.17, 17% Over the $83 IPO Price, as Nasdaq Fell 2%The $81.34 billion offering opened 34% higher, then gave back most of the gain while Nvidia, AMD and Micron dropped.
- 04Google's Gemini Tool Retrieval Preview Adds 26 Tokens Per Idle Tool and Publishes No Savings FigureAnthropic and OpenAI each reported token reductions for the same defer_loading flag. Google's November 19 post reports none.