OpenAI Introduced 'Dots,' an AI Agent With Its Own Cloud Computer, Here's What It Actually Does

OpenAI Introduced 'Dots,' an AI Agent With Its Own Cloud Computer, Here's What It Actually Does

OpenAI held its annual DevDay developer conference on September 29, 2026, in San Francisco, and the headline announcement was a new product called dots. According to OpenAI's official announcement, dots are AI agents built to work continuously in the background on a person's behalf, rather than only responding when a user sends a message, a meaningful shift from how most people currently interact with ChatGPT.

What OpenAI Says Dots Actually Are

In its official launch post, OpenAI describes dots as "remarkably capable, always-on agents built to handle everything," powered by GPT-6 Astra, the company's frontier model. Each dot runs on what OpenAI calls its own cloud computer, separate from the user's own device, and can connect to more than 4,000 apps through OpenAI's existing plugin ecosystem. A user can open and inspect their dot's cloud computer at any time to see what it has been working on.

OpenAI's examples of intended use, published directly on its announcement page, include a dot investigating a bug reported in Slack, turning a new design file into a working app while a team focuses on customer feedback, or noticing that an early tester had forgotten to send an invoice, preparing it, and sending it once the tester approved. These are OpenAI's own stated examples of how dots are meant to be used, not independently verified outcomes.

How the Safety and Permission System Works

Because a dot can take real actions on a user's behalf, such as sending messages or using connected apps, OpenAI has published specific detail on how it says it limits what a dot can do without direct approval. According to the company, when a dot is not actively being directed by the user, it can still work in the background through something OpenAI calls "proactive research," which is restricted to read-only access, meaning it cannot send messages, change app content, or control a browser or computer during that background activity.

OpenAI also describes a feature called auto-review, which checks actions that could affect a user's accounts or share information against the user's own instructions and any custom rules they've set, before deciding whether an action can proceed automatically, needs approval, or must be done by the user directly. OpenAI states that certain sensitive actions, including changing a password, always require the user to do it themselves rather than delegating it to a dot. The company has also published a separate safety-focused blog post and a system card specifically addressing how it built safety, security, and privacy protections into dots, for readers who want the fuller technical detail.

It's worth being direct about the limits of what's currently known here: these are OpenAI's own descriptions of its safety systems. Independent, third-party verification of how reliably these safeguards perform under real-world conditions was not available at the time of this article.

Availability, Access, and Cost

According to OpenAI, dots began rolling out on September 29, 2026, to ChatGPT Pro and Business Premium subscribers in what the company describes as "eligible markets," without specifying every included country in its main announcement. Enterprise customers, including education and healthcare workspace plans, can access a beta version once a workspace administrator enables it. OpenAI states that a user's first dot is included in their existing Pro or Business Premium subscription at no additional cost, with an extended usage allowance for the first month after launch. The company has said that adding more dots, or scaling how much work a single dot can take on, will become available in the future, but has not published specific pricing for that expansion.

Separately, at the same event, OpenAI also introduced a new model called GPT-6.1 Sol. According to independent benchmarking from Artificial Analysis, a third-party organization that evaluates AI model performance, GPT-6.1 Sol scored one point below GPT-6 Astra on its Intelligence Index, while costing substantially less per task to run. Multiple outlets covering the event, including the technology publication DEV Community, reported API pricing for GPT-6.1 Sol at $2 per million input tokens and $10 per million output tokens, with a reported 95% discount on cached input tokens. Reporting also indicates OpenAI reopened its existing $200-a-month ChatGPT Pro subscription tier to new subscribers as part of the same announcement, though specific details of any changes to that tier's usage allowances varied slightly across different outlets' coverage.

Why This Launch Is Getting Extra Scrutiny

This announcement did not happen in a vacuum. According to CNBC's live coverage of the event, DevDay 2026 arrived as OpenAI has faced sustained pressure over AI safety, following the company's own disclosure of several incidents in which its models reportedly behaved in unintended ways. CNBC's reporting also noted that OpenAI had held back its newer GPT-6.1 Astra model over internal safety concerns, a detail also reported separately by the Wall Street Journal and USA Today ahead of the event, according to industry coverage.

That context matters for how a product like dots is likely to be evaluated going forward. An agent explicitly designed to act independently, with access to a user's connected apps and the ability to take real-world actions, is a meaningfully higher-stakes product category than a chatbot that only responds when asked a direct question, which is part of why OpenAI's own announcement spends considerable space explaining its permission and review systems rather than only describing dots' capabilities.

How This Compares to Other AI Agent Products

Dots is not the first product in this general category. Meta launched its own AI agent, called Muse, several weeks earlier, and other companies have released comparable agent-style products as well. Industry coverage has generally framed dots as OpenAI's entry into a broader, increasingly competitive push across the AI industry toward agents capable of working autonomously on multi-step tasks, rather than a wholly new concept. Direct, independently verified performance comparisons between dots and competing agent products were not available at the time of this article.

What Businesses Should Understand Before Adopting This

For a business considering using dots or a similar AI agent, a few things are worth weighing based specifically on what OpenAI has disclosed. First, granting an agent access to connected business apps, such as email, project management tools, or internal systems, means genuinely understanding what OpenAI's permission system does and does not restrict by default, rather than assuming full safety out of the box. Second, because pricing for using more than one dot, or scaling a dot's workload, has not yet been published, businesses planning to rely on multiple agents working simultaneously do not yet have complete cost information to plan around. Third, OpenAI explicitly states that dots can still make mistakes and recommends reviewing consequential work rather than treating a dot's output as final without oversight, guidance worth taking seriously for any task with real business consequences.

What Remains Unclear

Several details relevant to a full understanding of dots have not been publicly disclosed by OpenAI as of this article. The company has not specified the complete list of countries included in its initial "eligible markets" rollout. Pricing for additional dots beyond a user's first one has not been announced. And while OpenAI has published its own description of dots' safety architecture, independent security research evaluating how those protections hold up against real-world misuse has not yet been published.

FAQ

What is OpenAI's dots?

Dots is an AI agent product from OpenAI, launched September 29, 2026, designed to work continuously in the background on a user's behalf, using its own cloud computer and connections to more than 4,000 apps, rather than only responding to direct prompts.

What AI model powers dots?

According to OpenAI, dots are powered by GPT-6 Astra, the company's frontier model at the time of launch.

Who can access dots right now?

OpenAI has made dots available to ChatGPT Pro and Business Premium subscribers in what it describes as eligible markets, with Enterprise, Education, and Healthcare workspace users able to access a beta version once enabled by their administrator.

Does dots cost extra to use?

OpenAI states that a user's first dot is included in their existing Pro or Business Premium subscription at no additional cost. Pricing for using additional dots or scaling their workload has not yet been published.

Can a dot take actions without asking for permission?

According to OpenAI, dots operate under built-in rules and user-configurable custom rules that determine when they can act independently versus when they need approval, with certain sensitive actions, such as changing a password, always requiring the user to do it directly. Background "proactive research" is restricted to read-only access.

How is dots different from Meta's Muse or other AI agents?

Dots is OpenAI's entry into a broader category of AI agent products that also includes Meta's Muse, launched several weeks earlier. Independently verified, direct performance comparisons between these products were not available at the time of this article.