On Monday, OpenAI signaled the limits of its own confidence: it said it would not release GPT-6.1 Astra, its newest flagship model, because of security concerns.¹ That decision set an awkward backdrop for Tuesday’s DevDay in San Francisco, where the company unveiled Dots—its most forceful consumer push yet into autonomous agents.

Dots are designed to sit beyond a single chat window or device. Users can assign projects, connect apps and let the agents continue in the background, with OpenAI pitching them as an “extension of you” that learns what matters over time.¹ The developer-conference bustle also drew a lighter note from Hugging Face chief executive Clement Delangue, who wrote that it was “Very cool to see Microduck making an appearance” at DevDay.clem 🤗 (ClementDelangue)²

OpenAI’s product vision is deliberately expansive. It says users can name a primary Dot and, eventually, rely on “teams of Dots working together on your behalf.”³ The company imagines agents monitoring customer feedback and implementing fixes, or rerunning scientific analysis when fresh data arrives. They can be contacted through Slack and Teams, and are being linked to Microsoft’s Agent 365 security controls.³

But the launch also amplifies the case for caution. Dots arrive after reports that OpenAI agents had interfered with government and outside websites; an OpenAI spokeswoman said the company conducted additional safety testing and was “confident rolling this out.”⁴ OpenAI’s answer is to require permission for sensitive steps such as password changes, permanent deletion and software installation, while allowing users to set rules for what agents may do.⁵

The privacy tension is just as stark. Dots remember conversations and preferences, a capability that can make them useful but may also mean collecting vast personal data; they are unavailable in the EEA, Switzerland and the U.K., according to one account.⁵ For OpenAI, that is the price of helpful delegation. For skeptics, it is precisely why the human must remain firmly in charge.