OpenAI Rolls Out Dots Agents Powered by GPT-6 Astra in ChatGPT

OpenAI on September 29, 2026 announced dots, always-on personal AI agents powered by its GPT-6 Astra model that run on their own cloud computers, beginning a gradual rollout in ChatGPT to Pro and Business Premium users in eligible markets and an admin-enabled Enterprise beta.

In its launch announcement, OpenAI described dots as agents that get to know what matters to a user, keep working on their behalf, and take important work off their plate. The company said dots learn from feedback over time, can work toward a user’s goals around the clock, and connect through its plugin ecosystem to more than 4,000 apps. Users start with a primary dot they can name, and OpenAI said it envisions teams of dots eventually working together on a user’s behalf.

Cloud Computer, Connected Apps, and Channels

Each dot works on its own cloud computer, with its own browser, alongside the apps a user has connected, and users can open the dot’s computer at any time to inspect its work. Users may also give a dot permission to connect to and use their own laptop; the getting-started guide states that local computer access starts turned off and is enabled through the ChatGPT desktop app with an Allow access confirmation.

OpenAI said a dot can run with one project even while working on several others. Its examples include dots that turn recurring customer feedback into tested pull requests, revise launch materials when scope changes, rerun a scientist’s analyses as new data arrives, update enterprise sales proposals as requirements shift, and prepare clips and show notes from interview transcripts. OpenAI recounted that an early tester’s dot noticed he had forgotten to invoice a publication, prepared the invoice, and sent it after his approval.

Dots can be messaged or called in ChatGPT on desktop, web, and mobile, and reached in Slack and Teams, with OpenAI saying they carry context across every channel. The launch post lists texting as coming soon; the help center describes it as a limited beta through a third-party provider, restricted to Pro users in the US and unavailable in Business or Enterprise workspaces. Dots cannot initiate calls to the user at launch.

Controls, Safeguards, and Data Policies

OpenAI detailed the guardrails in a companion safety, security, and privacy post. The company calls GPT-6 Astra its most capable and aligned model and said it is trained to refuse harmful requests, including biological or cybersecurity misuse. When a user is not actively working with a dot, the dot looks for ways to help through background tasks OpenAI calls proactive research; those tasks use read-only tools restricted in code, so they cannot send messages, change content in connected apps, or control a browser or computer, and any follow-up action must pass the usual rules and checks.

Before a dot takes actions such as sending emails or changing files, a separate system called Auto-review checks the planned steps against the user’s instructions, Custom Rules, and safety requirements. If Auto-review blocks a step, it returns the reason to the dot, which can ask for information or approval, try a permitted alternative, hand the step back, or stop. OpenAI said the controls enforcing Auto-review sit outside the environments dots can change, and that sandboxing restricts the code and tools inside each dot’s workspace.

Custom Rules let users allow specific actions, require approval, or block them, but cannot remove mandatory confirmations, handoffs, or core safety requirements. Changing a password or transferring money between financial accounts must be handed back to the user; permanently deleting data or installing unrecognized software requires confirmation each time; purchases with cards saved on merchant websites require approval. Secure sign-in pauses the model while the user enters credentials in a form sent directly to the browser environment, and saved-password flows use an encrypted credential service; OpenAI noted a secret pasted into a message or document may still be visible to the model.

Safety monitoring can pause a dot’s work and display a warning. The company’s privacy, security, and safety FAQ states that prompt-injection protections reduce but do not eliminate the risk of malicious instructions causing an unwanted action, and that dots can still make mistakes. The FAQ also states that dots are not yet available to users under 18, that a dot’s context can be deleted only by deleting the dot, and that individual dot memories cannot currently be viewed or edited. Activity View in the desktop app shows ongoing and delegated tasks, and users can redirect a dot or ask it to stop.

A dot receives memories and recent conversation context from ChatGPT, and conversations with a dot can contribute to ChatGPT memory; turning off the Memory setting stops that sharing but does not delete what the dot has already received. Limited human review may occur in safety-related cases even when model improvement is turned off.

On data use, OpenAI said it does not use content from ChatGPT Business, Enterprise, or Edu workspaces to train models by default, and that on personal plans the “Improve the model for everyone” setting controls whether a dot’s conversations and work may be used. Models are not trained directly on proactive research or its notes, though information brought from that research into an eligible conversation or task may be used depending on settings.

Availability, Plan Terms, and Specialist Dots

The rollout began on September 29, 2026 for Pro users in markets excluding the European Economic Area, Switzerland, and the UK, and for Business Premium users across all supported ChatGPT regions. Enterprise users, including Edu and Healthcare, can try the beta when a workspace admin enables it, and it is initially turned off by default. The rollout is gradual, and access may take several days to reach an account.

The first dot is included in Pro and Business Premium plans at no extra cost, with an allowance for deeper work. For the first month after launch, dots usage will not count toward eligible plan allowances; OpenAI said it will share usage terms for each plan afterward. Conversations with a dot do not count toward ChatGPT usage limits, while tasks a dot starts or manages in Codex or ChatGPT Work count as usual. OpenAI said users will later be able to add more dots and scale each dot’s output by increasing its speed or monthly work volume.

Dots are created in the ChatGPT desktop app or on desktop web and cannot currently be created on mobile or mobile web. A new dot’s default handle is @yourname-dot, which becomes @yourname-agentname once the dot is named. Users can schedule reminders and recurring checks, pause a dot from its profile, or reset it, which deletes the dot along with its conversations, saved memories, and scheduled tasks.

OpenAI is also previewing specialist dots that take on defined responsibilities inside organizations. The company said each specialist dot is set up with its own identity, credentials, and system access, with IT-provisioned hardware and deep integrations with company systems of record. OpenAI said it is building on early internal testing across procurement, invoice processing, email marketing, customer support, and commercial contracting, and is starting with focused enterprise pilots in which its engineers work directly with organizations to define each dot’s responsibilities, tools, and approval process. It said it is separately working with Microsoft to integrate specialist dots with the enterprise governance and security controls in Agent 365.

OpenAI said it plans to expand dots to more users soon and will share what it learns from everyday use to guide what it builds next.