October 10, 2026
Dots: OpenAI promises privacy for the agent, confidential computing expected in autumn
OpenAI opened Dots to Pro and Business Premium subscribers in ChatGPT in eligible markets from Tuesday, Sep 30, 2026. The first Dot is free for eligible subscribers; several Dots per user are still in the plans.
Dot works without a prompt and runs all the time. Each one has its own cloud computer with a browser and connected apps, more than 4,000 in total. You can start with your main Dot: give it a name and set it up for yourself. It talks through Slack and Microsoft Teams; SMS is promised later.
At DevDay, Sam Altman promised a new privacy standard for frontier AI, and OpenAI indirectly took a swipe at Meta Muse over how it protects user data.
Privacy works in two layers here. The first, Private Safety Processing, is automated safety checks with no access to the content for OpenAI staff. The second, Private Inference, is confidential computing. It is promised for the fall, and the announcement did not give a price.
Limits. Background tasks are read-only: the agent does not send messages or change content. Some actions require your confirmation, and the model cannot see password changes.
Assistants used to respond only while you were in the chat window. Dot works without a prompt, so the privacy question shifts to what the agent does without you. Alongside Dots, OpenAI announced a Pro 500 plan at $500 a month with limits 25 times higher than ChatGPT Plus.
Private Inference is promised for fall 2026, and OpenAI has not yet named a price.