OpenAI's Dots Work While You Sleep: What Do They Do Without Asking You?

OpenAI's Dots: An AI Agent That Works Around the Clock on Its Own Computer

At its DevDay developer conference on September 29, 2026, OpenAI announced dots: always-on AI agents powered by GPT-6 Astra. Each dot has its own cloud computer and browser, connects to more than 4,000 apps through OpenAI's plugin ecosystem, and can be reached from ChatGPT, Slack or Teams, by message or by voice call. The gradual rollout starts with Pro and Business Premium subscribers, and the first dot is included in those plans at no extra cost. The core difference from the ChatGPT you know is that a dot does not wait for your question: it keeps projects moving between conversations, researches in the background, and comes back with results and with the decisions that need you. So the most important question for a business owner is not what the agent can do, but what it will do without asking.

What is actually new compared with earlier assistants?

According to OpenAI's announcement and its official help page, four things set dots apart:

• A separate computer: the dot works on its own cloud computer, and you can open its screen at any time to see what it is doing. Connecting it to your own computer is optional and switched off by default.
• Continuous work on several projects: you can hand it a new task while it works on others, without a separate thread for each project and without directing every step.
• Proactive research: when you are not talking to it, it reads the apps you have connected, looking for ways to help, and builds memories from them. OpenAI says the tools for this research are read-only and that this is enforced in code: they cannot send messages to anyone, change content in connected apps, or control a browser or computer.
• Scheduled tasks: for example, reviewing your calendar every morning and telling you what is coming up.

One of OpenAI's own examples: an early tester's dot noticed he had forgotten to invoice a publication, prepared the invoice, and sent it after his approval. That example sums up the whole idea: the agent catches forgotten work, but the send still went through its owner's approval.

What does it do without asking, and what needs your approval?

Dots start with built-in rules, and you add Custom Rules on top. For each action you pick one of four behaviors: take action without asking, take action if pre-approved (meaning you explicitly requested it in your prompt), ask before taking action, or hand off to you. Some limits are fixed, and your rules cannot remove them:

• Your confirmation every time: permanently deleting data, installing or running software from an unrecognized source, and granting new security-sensitive access.
• Always handed back to you: changing a password and transferring money between financial accounts. The dot helps with the surrounding task, but you take that step yourself.
• Purchases: it can buy using cards already saved on a merchant's website, but every purchase requires your approval.
• Sharing data: the more sensitive the information, the more precisely you must name the recipient. Health data, for example, always requires a named recipient.

Before actions such as sending an email or changing a file, a separate system called Auto-review checks the planned step against your instructions, Custom Rules and safety requirements, for instance the recipient and content of an email. OpenAI says the controls that enforce this check sit outside the environments the dot can change, so the dot cannot switch it off. Even so, the company states plainly that dots can still make mistakes and advises reviewing any consequential work.

The detail many will miss: disconnecting an app does not erase what the dot learned

OpenAI's help page states that disconnecting an app stops new information reaching the dot but does not delete information it already obtained from that app. To remove it, you have to reset the dot, which deletes it entirely along with its conversations, saved memories and scheduled tasks. In practice: if you connect company email or your customer system on the first day of a trial, everything the dot read stays in its memory until you delete the whole dot. Since your customers' data is governed by the Personal Data Protection Law, the decision about what to connect has to come before the trial, not after it. We covered the obligations in your customer conversations and the PDPL.

Company data: a personal plan is not a business workspace

OpenAI says it does not use content from ChatGPT Business, Enterprise or Edu workspaces to train its models by default. On personal plans such as Pro, you control this through the "Improve the model for everyone" setting, and when it is on, training can include actions the dots take, after the company works to remove personal identifiers. So if your staff will try dots on company data, do it from a workspace the company manages, not from scattered personal subscriptions whose settings you cannot see.

Is it available in Saudi Arabia, and what does it cost?

For Pro subscribers, the rollout excludes only the European Economic Area, Switzerland and the United Kingdom, and Saudi Arabia is not among them. For Business Premium it is available in all supported ChatGPT regions. Enterprise workspaces, including Edu and Healthcare, can try the beta once a workspace admin enables it; it is off by default. The rollout is gradual and may take several days to reach your account. You create the dot in the ChatGPT desktop app or in ChatGPT on desktop web, then talk to it from the mobile app. Texting is a limited beta, currently for Pro users in the US only.

On cost, the first dot is included in Pro and Business Premium at no extra charge, with an allowance for deeper work and extended limits for the first month after launch. Conversations with your dot do not count toward ChatGPT usage limits, but tasks it starts in Codex or ChatGPT Work count as usual. OpenAI says you will later be able to add more dots and raise each one's speed or the amount of work it takes on per month, but it has not published prices for that yet.

Specialist dots: the next step inside companies

Beyond the personal dot, OpenAI is previewing specialist dots that a company sets up with their own identity, credentials and access to the systems they need to take on a defined responsibility. OpenAI says it is building on internal testing across procurement, invoice processing, email marketing, customer support and commercial contracting, that it is starting with focused enterprise pilots, and that it is working with Microsoft to manage specialist dots through Agent 365. This raises the real question for any business: are your internal systems ready for an agent with its own identity? We covered that in AI agent identity and how it should log into your systems safely.

Before you switch on your company's first dot

• Start with low-sensitivity apps: connect one or two, such as a calendar and a general documents folder, before email and your customer system.
• Write the rules before the tasks: state explicitly: never email a customer without asking me, never share a file outside the company.
• Assign a reviewer: Activity View shows ongoing and delegated tasks, so make reviewing it part of someone's day.
• Keep the record in your own systems: anything the agent does in your invoicing or inventory system should appear in your audit log under its own name, not the employee's.
• Measure before you expand: run it for two weeks on one recurring task, such as a weekly report, then judge its accuracy and the time it saved. We explained how in how to test an AI agent before trusting it with operations.

How we build for it at Origami

However capable an agent is, it can only work as well as your systems allow. When it reads from a shared spreadsheet or an open mailbox, it is hard to control what it sees and what it changes. That is why we build our clients APIs with specific permissions for each user and each agent, an audit log that shows who did what and when, and approval paths an outside tool cannot bypass; we explained the foundation in connecting your business systems through APIs. That way you can use dots today, or any other agent tomorrow, and your data and decisions stay under your control. For the wider picture, see securing and governing AI agents.

Sources

• OpenAI: Introducing dots (September 29, 2026)
• OpenAI: How we build safety, security, and privacy into dots
• OpenAI Help Center: Getting started with your dot
• SDAIA: Personal Data Protection Law