OpenAI Dots: What Can the Always-On AI Access, and How Do You Keep Control?
OpenAI’s new Dots can work in the background across the apps you choose to connect. They do not automatically gain access to your personal computer. The important distinction is between read-only background research and a task you explicitly ask a Dot to carry out. Here is what to check before giving an always-on assistant access to your work and personal information.
What are OpenAI Dots?
Announced on September 29, 2026, Dots are AI assistants designed to pursue ongoing goals rather than wait for a single prompt. OpenAI says a Dot can use its own cloud computer and browser, connect to supported apps through plugins, research a topic and report back as work develops. The system uses GPT-6 Astra.
That can be useful when a goal spans several services: following a project, collecting relevant updates or preparing a summary. It also makes the scope of access a practical decision. A Dot can only draw on the information and tools made available to it, so the apps you connect and the instructions you give matter.
Can a Dot see my phone or laptop automatically?
No. OpenAI describes a Dot’s cloud computer as separate from your personal device. Connecting an app may expose information from that app according to the connection’s permissions, but creating a Dot does not itself give it control of your own laptop or phone.
Before connecting any service, look at the specific account and data the connection covers. A calendar, inbox, file store and messaging app each contain different kinds of information. Give access to the smallest useful set for the goal, and revisit that choice when the goal changes.
Can it send messages or change files while I am away?
OpenAI distinguishes between two modes. Its proactive research in the background uses connected app tools in a read-only way. In that mode, it cannot send messages, alter content or control a browser or computer.
A Dot can also perform a task you ask it to do, subject to the tools, rules and approval checks that apply to that action. OpenAI says consequential actions may go through automatic review, while some actions require the user. Do not assume that every action will be blocked until you personally confirm it: inspect the available controls and the permissions of each connection.
How do I keep control of a Dot?
- Start with one narrow goal. A clear assignment makes it easier to judge whether an app connection is necessary.
- Review connected apps. Connect only the services the assignment needs, and check the account and permission scope.
- Set Custom Rules. Use the available rules to allow or restrict actions that matter to you.
- Check the activity view. Inspect what the Dot has researched and done, especially after a new connection or instruction.
- Remove access you no longer need. Revisit connections as projects end or change.
This is a better habit than relying on a broad request such as “handle everything.” It keeps the task understandable and lets you spot an incorrect assumption before it spreads into later work.
What happens to my data?
The answer depends on the account and its settings. OpenAI says personal-plan users have a control for whether their content helps improve models. It says Business, Enterprise and Edu content is not used for model training by default, and that proactive research and notes are not trained on directly. Those are separate questions from whether a connected app’s content is available to a Dot for the task you assigned.
If a connection includes sensitive material, check its access before switching on an ongoing goal. No assistant is infallible: OpenAI itself says Dots can make mistakes. Review important findings and actions against the underlying source.
Who can use Dots now?
OpenAI says Dots are rolling out to eligible Pro and Business Premium users in supported markets, with an Enterprise beta controlled by administrators. The first Dot is included in eligible plans without an added charge, with an allowance for deeper work. Availability and limits can vary by plan and market.
The practical takeaway: an always-on AI is most helpful when its job and access are specific. Before creating one, decide what result you want, which app actually contains the needed information and what actions you are comfortable allowing. Then check its activity as it works.
