An artificial intelligence agent can already search for information, write code or complete a form. OpenAI now tries to give it a long-term responsibility: to monitor a project, to resume work when new data arrives and to return to the user only when a decision is needed.

Launched on September 29, 2026, Dots transforms ChatGPT into a persistent assistant. The product is actually being deployed, but its autonomy remains under control, and its capabilities have not yet been independently evaluated on a large scale.

A cloud computer for each agent

Each dot has its own cloud computer, a browser, and access to applications that the user chooses to connect. It works with GPT-6 Astra and can continue several tasks between conversations, according to the Official OpenAI presentation.

The idea is to exceed the one-time request. A dowry could monitor comments received by an application, prepare corrections or restart an analysis when new data arrive.

OpenAI says its ecosystem can connect Dots to more than 4,000 applications. This figure comes from the manufacturer; it does not mean that each connector allows the same actions or that it is available in all packages and countries.

The user can join his agent in ChatGPT, Slack or Teams. Text messaging is announced, but its initial beta version is limited to US Pro users.

# Read in the background, act under conditions

When no one gives him a new directive, Dots can launch a "proactive search" in already authorized sources. OpenAI states that this work uses read-only tools: it cannot directly send an email, modify an application, or control a browser or computer.

Further action requires rules. The user may authorize certain operations, require approval or block them. A second system, Auto-review, checks the recipient and the information that an agent wants to share before execution.

Some steps remain reserved for the person. The OpenAI Security Document indicates that a transfer between bank accounts or a change of password must be given to the user. A purchase with a card already registered requires approval.

These barriers reduce the risk without removing it. OpenAI recognizes that Dots can be wrong and recommends checking important work. During the launch demonstration, some trials also experienced failures, reported Reuters and Axios.

Memory, Access and Data Control

A persistent agent becomes more useful by accumulating context, but also more sensitive. Dots can form memories from connected applications. Unplug application cuts new accesses, without automatically deleting information already learned.

OpenAI states that the content of Business, Enterprise and Edu spaces is not used by default to improve its models. In a personal package, the "Improving the Model for All" setting determines the possible use of Dots' conversations and work.

A compatible password can be transmitted to the browser without entering into the context of the template. This protection does not cover a secret placed in a document legible by the agent.

The most difficult risk remains the injection of instructions: a web page or email may contain a malicious order to divert the agent. OpenAI combines restrictions, monitoring and approvals, but company agents have recently been associated with unauthorized access during evaluations, according to a TechCrunch survey. The new product will have to demonstrate that its safeguards are resistant to daily use.

Prices and availability in Canada

Dots is gradually deployed in ChatGPT. The first agent is included without charge in the Pro and Business Premium packages; Enterprise, Edu and Healthcare administrators can activate a beta version. Access can take several days.

The Official Help Sheet initially excludes the Pro package of the European Economic Area, Switzerland and the United Kingdom. It does not exclude Canada, but availability always depends on the account, package and progressive deployment.

OpenAI has published no standalone prices for Dots, nor tariff to add other agents, nor Canadian cost specific to this function. The total may vary depending on the package, taxes, limits, integrations and region.

Limits or Considerations

An agent who works for a long time may add to a bad statement before a person notices it. Minimum permissions, trial records, explicit approvals and review of the activity log remain essential.

Organizations will have to verify the residence and retention of data, third party connectors and contractual responsibilities. A task performed by Codex or ChatGPT Work continues to use their quotas.

Finally, the scenarios presented by OpenAI are examples and statements by the manufacturer. They do not yet prove that Dots reduces delays, errors or costs in a real business.

Conclusion

Dots is less a new chatbot than an attempt to give an agent a lasting mandate, memory and workspace.

At the same time, this autonomy increases the amount of data available and the scope of an error. The success of the product will therefore depend as much on the quality of the controls, newspapers and approvals as on the intelligence of the model.