Amid the highly effective operations of Meta’s Muse personal agent and SpaceXAI’s Grok Bot, OpenAI has officially unveiled Dots, a background virtual assistant system that operates continuously 24/7. Powered by GPT-6 Astra, each Dot possesses its own dedicated browser and cloud computer, while also maintaining compatibility with more than 4,000 applications across OpenAI’s plugin ecosystem.

The core distinction between Dots and traditional chatbots is its ability to proactively initiate tasks. OpenAI provided concrete examples: if it notices a dinner appointment on your schedule, Dot will automatically send two food recommendations along with prices and proceed with the reservation only after you make a selection. For developers, Dot can monitor user feedback, handle minor bugs on its own, run builds, test, and submit PRs complete with video demonstrations of the changes.

Not yet for everyone

OpenAI has begun rolling out Dots exclusively to ChatGPT Pro, Business Premium, and Enterprise accounts in eligible regions. The company confirmed that Dots does not yet support the standard ChatGPT tier. Additionally, interacting with Dot will not consume usage limits, except when assigning tasks to it within ChatGPT Work or Codex.

At this stage, users face a key limitation: each account can only create a single Dot. However, OpenAI plans to add cloning capabilities in the future. Mentions of scaling up compute capacity or expanding monthly workloads for each Dot in their announcement point toward a future performance-based pricing model. Beyond the ChatGPT environment, Dot also integrates with Microsoft Teams and Slack to maintain seamless context across multiple devices, while Pro users can join a waitlist for RCS and iMessage messaging.

Specifically for enterprise clients, OpenAI launched a preview of specialist dots. These agents have distinct identities, unique permissions, company IT-provisioned hardware, and deep integration with internal data systems. At DevDay, Altman also showcased specialized versions designed for legal analysis, email marketing, and accounting.

Questions that need to be asked

Despite being equipped with guardrails, security risks remain. OpenAI’s announcement does not emphasize that Dots requires explicit user approval before carrying out sensitive actions (such as installing software or changing passwords). The system also offers Custom Rules to define boundaries and an auto-review feature to assess compliance. Nevertheless, having an agent proactively and continuously scanning the web and sending messages still introduces attack vectors.

Furthermore, data privacy is a crucial aspect to note: the policy of training models on personal data still applies to all interactions with the agent. If your account has data sharing for model training enabled, all communications with Dot will fall under this scope as well. For a tool directly connected to email, work calendars, and Slack, this is a term that warrants careful consideration.

The AI agent market is also witnessing fierce competition alongside numerous safety challenges. Meta’s Muse app topped mobile download charts within weeks of launch, but Meta immediately encountered a security incident: a user reported that Muse shared their private home address with a buyer on Facebook Marketplace without permission. In addition, WIRED noted that Meta previously had to patch a zero-day vulnerability in this agent.

The greatest challenge today lies not in technological limitations, but in trust: no organization has yet demonstrated that a continuously running agent can ensure absolute reliability in small tasks without human verification. OpenAI’s temporary restriction of one Dot per account reflects a cautious approach. The open question for the road ahead is when users will be allowed to run multiple agents simultaneously, and what governance mechanisms will be in place.