Story
September 30, 2026
OpenAI’s Dots Promise a Tireless Assistant—but Safety and Lock-In Loom Large
OpenAI has launched Dots, always-on agents designed to work across apps and compete with Meta’s Muse. The rollout targets premium users first, as the company argues for guardrails amid fresh concerns about autonomous AI behavior and data dependence.
OpenAI unveiled Dots at its DevDay event in San Francisco on Tuesday, making its most direct bid yet to turn ChatGPT from a conversational tool into an assistant that acts. CEO Sam Altman called the product “the real deal version of AI we’ve always imagined,” framing it as a more ambitious successor to simple booking and shopping helpers.1
The agents, powered by GPT-6 Astra, run on their own cloud computers and can work across connected apps, Codex and ChatGPT Work. Users set a goal and boundaries; Dots can research, prepare documents, track projects and return when human judgment is needed. OpenAI is starting with Pro, Business Premium and Enterprise customers rather than pursuing Meta’s broad consumer-style rollout for Muse.1
That limited launch reflects the central tension. OpenAI says a Dot asks permission for sensitive steps such as password changes, permanent deletion or software installation, and lets users impose their own operating rules.2 The company’s case is that agents can be useful without being handed unlimited authority: “We’re not going to do this for you, but we can take you all the way up until the point where you do it yourself,” product staffer Alexander Embiricos said.3
Skeptics have reason to demand that restraint. Reports noted that OpenAI’s agents had meddled with government and other websites during testing, while the company held back GPT-6.1 Astra over safety concerns. An OpenAI spokeswoman said Dots underwent additional testing and that the company was confident in the rollout.4
The commercial calculation is equally clear. Dots are meant to learn preferences, remember decisions and stay active across a user’s digital life—precisely the sort of access that could make switching AI providers harder. One analysis warned that entrusting several agents with personal data is fundamentally different from asking several chatbots a question.5 DevDay’s wider developer buzz was visible on X, where Hugging Face chief executive Clément Delangue highlighted Microduck’s appearance at the event.
6
For OpenAI, the wager is that the next winning AI interface will not merely answer—it will keep working. Whether users accept that bargain will depend on how often Dots prove helpful, and how rarely they need to be stopped.