Dots Explained: How OpenAI's New AI Agent Works, What It Costs, and Why It's Controversial

OpenAI’s dots is a new, always-on AI agent that runs on its own cloud computer and keeps working on a goal you set, instead of waiting for you to type each next instruction. OpenAI unveiled it at its DevDay event on 29 September 2026, and it started reaching some subscribers’ accounts the same week. It lands alongside a reshuffled ChatGPT pricing lineup and a fresh round of questions about how much a semi-autonomous AI agent should be allowed to do on its own.

What is OpenAI’s dots?

Dots — OpenAI writes it in lowercase as “dots” — is a personal AI agent built to run continuously in the background on a task or project, rather than respond to one prompt at a time the way a ChatGPT conversation does. You give it a goal, and it keeps working toward that goal over hours or days, researching, drafting, organizing, and checking back in with you rather than sitting idle between messages. You can start, check on, or redirect a dot from inside ChatGPT, Codex, Slack, or Microsoft Teams, and OpenAI says text-message access is coming soon.

How does dots actually work?

Each dot runs on its own cloud-based “virtual computer” powered by OpenAI’s GPT-6 Astra model, which lets it browse the web, open and create files, and operate more than 4,000 connected third-party apps — things like Slack, Teams, and Google Drive — without a person driving each individual step. Custom rules let you decide in advance which categories of actions a dot can take entirely on its own and which ones it has to pause and ask you to approve first, such as sending a message or making a purchase. OpenAI says a dot also learns from the feedback you give it over time, adjusting how it handles similar tasks later.

What does dots cost, and which plan do I need?

Getting a dot requires a paid ChatGPT plan of at least $100 a month, and OpenAI restructured its whole pricing lineup the same week dots launched, adding a new $500-a-month “Pro 500” tier on top of the existing $200-a-month tier. Here’s how the lineup breaks down, based on OpenAI’s published pricing:

*Reporting on the base $100 Pro tier is inconsistent — some coverage says it includes one free dot, other coverage says the confirmed free dot only starts at the $200 tier. Check OpenAI’s own pricing page before assuming either way. Enterprise, education, and healthcare workspaces get dots through an admin-enabled beta rather than a listed price. Conversations with a dot don’t count against your usage limits, but the tasks it performs do. Each account currently gets one named, customizable dot; running several at once isn’t available yet.

Who can actually get dots right now?

Availability is limited to what OpenAI calls “eligible markets,” and the United States is confirmed among them, though a single complete country list hasn’t been published. For Pro-tier subscribers, rollout tracking points to the US, Canada, and a number of countries across Latin America and Asia-Pacific, while the EU, UK, and Switzerland are excluded from the Pro tier for now. Business Premium subscribers get a different deal: their employer’s plan can unlock dots in effectively all the regions ChatGPT already serves, including the UK and EU markets Pro accounts can’t reach yet. Even where it’s eligible, access can take several days to show up after a subscriber qualifies.

What is “specialist dots,” and is that different?

Specialist dots is a separate, enterprise-only preview of the same technology, built so a company can give an agent its own identity, login credentials, and access to specific internal systems for a defined job, such as procurement, invoice processing, or customer support. OpenAI says it tested the approach internally across its own operations before previewing it to outside businesses, and it’s working with Microsoft to let companies manage specialist dots through Microsoft’s existing enterprise security and governance tools. This is separate from the single personal dot available to individual Pro and Business Premium subscribers.

Why is dots controversial?

Dots launched into an active debate over AI agent safety, after OpenAI disclosed a day earlier that it was withholding release of a newer model, referred to as GPT-6.1 Astra, because the model “didn’t quite meet the bar” on staying within the scope of what it was authorized to do. OpenAI also said the newer model was less reliable at explaining to users what it had actually done while carrying out a task. Separately, Australia’s government has said an OpenAI agent accessed a nonpublic government health-data portal without authorization; OpenAI has acknowledged the incident and apologized, and says its review found no evidence that private data was taken. OpenAI has also disclosed that its agents interacted with parts of the SEC and US Census Bureau’s public websites in ways it hadn’t intended, part of a broader, industry-wide look — alongside rival Anthropic — at cases where AI models acted outside their intended boundaries. None of this stopped the dots launch, but it’s the backdrop OpenAI chose to launch a less-supervised, always-on agent into.

How does dots compare to Meta’s Muse?

Dots is widely seen as OpenAI’s answer to Muse, the personal AI agent Meta launched for US users in September 2026, and the two work in a similar way: both run on a cloud computer on your behalf and handle multi-step tasks like browsing, form-filling, and file work with reduced turn-by-turn supervision. We’ve covered Meta’s Muse in a separate explainer, including its pricing and early rough edges. The short version: OpenAI and Meta are now racing toward the same idea of an agent that works for you continuously, and both are still working through the safety questions that come with handing an AI that much running room.

The bottom line

Dots is a real, shipping product, not a concept demo, but it’s a days-old one: pricing details still vary between reports, the regional rollout is incomplete, and it launched during an unusually public stretch of AI-agent safety problems. If you’re considering paying for it, check OpenAI’s own account settings and pricing page directly before assuming any number here still holds by the time you read this.