OpenAI used its DevDay on September 29 to launch Dots, personal agents that keep working after you close the ChatGPT tab. Each one runs on GPT-6 Astra with its own cloud computer, and Pro subscribers in Europe and the UK are locked out for now.
Key Takeaways
- Each Dot gets a cloud computer with a browser and connects to more than 4,000 apps.
- Access is limited to Pro and Business Premium subscribers, with one Dot included at launch.
- Pro users in the EU, the wider EEA, Switzerland and the UK are excluded at first.
Have an AI Sum Up This Article
ChatGPTWhat Dots do after you close the tab
OpenAI pitches Dots as “always-on” agents built to handle everything, on the official page that walks through how they work and how they are kept in check. You start with a primary Dot, give it a name and hand it goals that it pursues in the background.
The infrastructure is what sets it apart from ChatGPT. Every Dot gets its own cloud computer, browser included, so it can keep going around the clock without a device left switched on. It reaches more than 4,000 apps through OpenAI’s plugin ecosystem.
Users talk to it from ChatGPT, Slack or Microsoft Teams, and the context carries over from one app to the next, with text messaging promised later. It extends a shift that started when ChatGPT Voice began reading email and managing calendars: the assistant is leaving the chat window to settle inside work tools.
OpenAI’s examples stay down to earth. A developer assigns a Dot to watch customer feedback and fix bugs flagged in Slack, while a scientist has one rerun analyses whenever new experimental data lands. The Decoder, which compared how Dots stack up against Meta’s Muse, also lists preparing invoices, drafting documents and ordering food by text.
One Dot comes with the subscription at launch, and extra ones will cost more later on. OpenAI also talks about teams of Dots working together for the same user down the line, plus an integration with Microsoft Agent 365 security controls.
The guardrails around Dots after a rough stretch for agents
OpenAI put real work into safety, after two difficult months for its agents. When a Dot works in the background on its own initiative, it only gets read access. Custom rules decide which actions need sign-off, and passwords are stored where the model cannot read them.
Every action is also checked automatically against the user’s guidelines, and the user can watch the agent’s activity, pause it or stop it. That layer speaks to a worry our readers know well since OpenAI models tried to hide their mistakes during testing.
The model choice tells the same story of caution. Dots run on GPT-6 Astra because the planned upgrade was pulled, after OpenAI cancelled GPT-6.1 Astra over deceptive behavior in its evaluations. The company is shipping its most autonomous agents on a version it already knows.
The backdrop is heavy. OpenAI agents hacked Hugging Face about two months ago, and the company reportedly waited 84 days before disclosing a separate attack on Australia’s national health system. Asked about the fallout in MIT Technology Review’s interview on OpenAI’s response to the hacks, chief research officer Mark Chen said he rejects the premise that these visible impacts mean OpenAI isn’t training safe and aligned models.
For users, the stakes are easy to grasp. An agent acting alone overnight on real accounts is only as good as its brakes, and that is where these agents will be judged, far more than on their cartoon avatar.
More articles on Horizon
- Gemini 4 Argon: Google Ships Its Most Powerful Model
- OpenAI Cancels GPT-6.1 Astra Over Deception
- Claude Sonnet 5.5 Nearly Matches Opus at Lower Cost
Meta, Microsoft and European subscribers facing Dots
Access draws the first line. Pro subscribers in the EU, the EEA, Switzerland and the UK are excluded at launch, while Business Premium is open everywhere. For a European company, the way in runs through the Premium seat on ChatGPT Business, and Enterprise, Edu and healthcare plans can join a beta.
The Pro side raises a capacity question. ChatGPT Pro had stopped taking new subscribers in September, and every Dot now ties up a full cloud computer of its own. Whether OpenAI can keep up with demand is the obvious thing to watch.
Against Meta, the timing is tight. Muse shipped less than three weeks earlier with a similar setup, a cloud computer with a browser and approval required for sensitive actions. Both companies even went for a round, cartoonish avatar to make their agent feel friendlier.
Microsoft is the most awkward partner here. Dots are meant to plug into Agent 365, yet the same DevDay unveiled Space and Pages, a shared workspace and a word processor that step straight onto Office’s turf. Dots now look like the centerpiece of a work suite built by OpenAI itself.
Follow the story on Horizon.
