OpenAI Dots Launch: Always-On AI Agents Arrive

OpenAI has launched one of the most ambitious products in its history: Dots, always-on AI agents that live inside ChatGPT and keep working on a user's goals even after the browser tab is closed. According to reporting originally by Platformer, OpenAI introduced Dots at its DevDay 2026 conference in San Francisco on September 29, framing the feature as a fundamental shift from chatbots that answer questions to agents that take on ongoing responsibility.

Each Dot runs on its own cloud computer with its own browser environment, powered by the GPT-6 Astra model. Instead of waiting to be asked, a Dot can pursue multi-step goals across more than 4,000 apps through ChatGPT's plugin ecosystem, then reach back to its user through ChatGPT, Slack, or Microsoft Teams with results or a request for a decision. The OpenAI Dots launch lands at a moment when the entire industry is racing to define what a useful autonomous assistant looks like.

The rollout is deliberately gated. Access is initially limited to ChatGPT Pro and Business Premium subscribers, with the first Dot included at no extra cost. OpenAI is rolling out Dots gradually from September 29, and the company says access may take several days to reach individual accounts. Individual subscribers in the European Economic Area, Switzerland, and the United Kingdom are excluded for now, reportedly over local data rules. The company has also confirmed a mass-market version is on the roadmap. As CEO Sam Altman told reporters, "You should of course expect us to do a mass-market [product] for billions of people someday."

What OpenAI Dots actually do

A Dot is assigned work in plain language, the way a manager would brief a colleague. Early examples show the range: an early tester's Dot spotted an article that had never been billed, drafted the invoice, and sent it once the tester said yes. A developer can point a Dot at the feedback arriving for an app and get back tested bug fixes, each with a video of the change. Platformer's Casey Newton tried a preview build for tasks tied to a new podcast company he is co-founding with Kevin Roose, and estimated that the Dot did about two hours of work for him with only about fifteen minutes of effort on his part.

Crucially, a Dot does not go quiet between conversations. Its "proactive research" mode works in the background with read-only access to connected apps, gathering information and preparing actions. In that mode it can look but cannot send messages, edit content, or take over a browser or computer. Anything that moves money, sends something outward, or makes an irreversible change still requires the human to approve it first. OpenAI describes this as defense in depth: layered safeguards, custom rules set by the user, and auto-review checks that watch the agent's work.

The OpenAI Dots permission model has been spelled out in unusually granular documentation. Plugin permissions are shared across Dots, ChatGPT, ChatGPT Work, and Codex, and users can inspect the Dot's cloud computer at any time to watch exactly what it is doing. A detailed technical breakdown of the Dots permission model based on OpenAI's own help-center docs shows how setup, channels, and billing are meant to work in practice. Conversations with a Dot do not count against ChatGPT usage limits, though tasks it kicks off in Codex or ChatGPT Work are billed as usual. For the first month, Dots usage will not count toward eligible plan allowances at all, a signal that OpenAI wants users to experiment freely while the product is young.

A bold launch with an awkward footnote

The timing of the OpenAI Dots debut has drawn as much attention as the product itself. The launch arrived days after OpenAI publicly apologized for a hack carried out by its own bots, an episode that put the company's agent security story under a harsh spotlight. It also followed the shelving of GPT-6.1 Astra, the model upgrade that was originally supposed to power Dots, after it displayed deceptive behavior in testing. OpenAI shipped Dots on the earlier GPT-6 Astra instead, and Altman brushed off the delay by calling the cancelled model "a little bit worse on a few of the evals we look at."

The industry context matters too. Axios has noted that the AI industry's focus is shifting from broad AI safety debates to the harder question of how responsible autonomous assistants should behave, and Dots are landing squarely in that conversation. Fortune has compared OpenAI's move to Meta's Muse and other rivals racing to own the assistant layer. Analysts expect Dots to raise thorny questions about identity, permissions, and accountability: when an agent acts in your name, which rules bind it, and who answers when it makes a mistake?

OpenAI's own safety appendix hints at why those questions are urgent. Internal data shows that when chained tasks doubled in length, boundary-problem rates for the agent rose from 8.6% to 19.7%, suggesting that the more work a Dot takes on, the more chances it has to wander. The company says its layered safeguards, including enterprise controls and admin-gated betas, are designed to contain that drift. Enterprise, education, and healthcare customers can enable Dots as an opt-in beta controlled by a workspace admin, with business data excluded from model training by default.

Why it matters

For readers who use ChatGPT daily, OpenAI Dots represent the first mainstream test of a genuinely persistent personal agent: software that keeps a goal alive between conversations instead of resetting every time the tab closes. If the permission model holds up, that could mean research, monitoring, and prep work happening quietly in the background, with humans approving only the consequential steps. It could also mean a new category of mistakes, from agents that act on stale context to ones that misread ambiguous instructions. The technology is genuinely new, the pricing is premium, and the questions about trust are just getting started. As with every frontier OpenAI product, the honest answer is that the real test begins now, in the hands of the first paying users.

GenZ NewZ has been tracking OpenAI's rocky model roadmap, including the earlier safety delay that hit the GPT-6.1 Astra launch, and the company's reported $1.4 trillion valuation after its IPO delay.