OpenAI's launch page includes the real Dots product demonstrations. Because the available guidance does not clearly authorize redistribution of those specific images outside the original page, AUXER keeps the visuals at the source.
View OpenAI's official images and demonstrations ↗Visual source: OpenAI, September 29, 2026. AUXER has not copied or re-hosted the launch imagery.
OpenAI introduced Dots on September 29, 2026, describing them as AI agents designed to keep working on a user's behalf. The difference from a conventional chatbot is easy to describe and much harder to engineer: the assignment can continue after the conversation ends.
According to OpenAI, each Dot is powered by GPT‑6 Astra, has its own cloud computer, and can connect to more than 4,000 applications through the plugin ecosystem. Users can talk to their Dot in ChatGPT, Slack or Teams, inspect ongoing work, and provide new direction while tasks continue to run.
What changes compared with today's agents?
The core idea is persistence. A typical agent receives a task, executes a workflow, and finishes. A Dot can maintain context about goals, preferences and quality standards, track changes over time, and continue work without requiring the user to direct every individual step.
OpenAI gives examples such as monitoring customer feedback and preparing software fixes, revising launch material when a product's scope changes, updating research analysis as new evidence arrives, and adapting sales proposals during enterprise deals. These are vendor-provided examples, not guarantees that every workflow will complete correctly without supervision.
A dedicated cloud computer and background work
Each Dot operates inside its own cloud computer. OpenAI says users can open that environment to inspect work and may optionally allow the Dot to connect to other devices, including a personal computer.
Dots also include a feature called proactive research. When a user is not actively working with the agent, it can look for potentially useful information in permitted connected sources. OpenAI says these background research tools are restricted to read-only access: they cannot directly send messages, change app content, or control a browser or desktop.
Permissions become a core product feature
Persistent agents increase the chance that a mistaken action can leave the chat environment and affect real systems. OpenAI therefore describes a design that combines app permissions, user-configurable rules and a separate review layer for certain actions.
- Custom Rules: users can allow, block or require approval for specific categories of action.
- Auto-review: a separate safety system checks actions such as sending messages or changing files before execution.
- Protected credentials: supported saved-password flows can sign in without placing the password in the model's context.
- Sensitive actions: some actions require confirmation, while steps such as changing a password or transferring money between financial accounts must be handed back to the user.
OpenAI explicitly notes that Dots can still make mistakes. That caveat matters because the value of a persistent agent depends not only on capability but also on containing the consequences of misunderstandings and malicious content such as prompt injection.
Who gets access first?
The rollout begins with Pro and Business Premium users in eligible markets. OpenAI says Enterprise users, including Edu and Healthcare, can try the beta when a workspace administrator enables it. The company is also previewing specialist Dots with their own identities and credentials for well-defined responsibilities inside organizations.
OpenAI also announced work with Microsoft to integrate specialist Dots into Agent 365, allowing companies to apply enterprise identity, governance and security controls around them.
Why this launch matters
The shift is not simply a smarter model. Dots push AI toward systems that maintain responsibility over time. That makes them closer to roles that today require continuous human attention: monitoring change, deciding when to act, preparing work and escalating decisions when approval is required.
For developers and businesses, the question changes from “can the model complete this task?” to “how long can it pursue this goal, with which permissions, and how do we know when it should stop?” Persistence turns memory, identity, auditing and access control into product features rather than infrastructure details.
What to evaluate before deploying an always-on agent
In real environments, teams should begin with minimal access and expand permissions only after validating behavior. They also need clear definitions for irreversible actions, human approval points, context-retention policies and auditability.
Dots are one of the clearest examples yet of the move from conversational AI to operational AI. The launch shows where the industry is heading. How much autonomy proves reliable in everyday use will need to be measured outside launch demonstrations.
Official sources
- OpenAI: Introducing dots, product announcement published September 29, 2026.
- OpenAI: How we build safety, security, and privacy into dots, covering isolation, permissions, Auto-review and proactive research.
- OpenAI Brand Guidelines, reviewed to assess permitted use of visual and brand assets.