Image: OpenAI

Some early users of ChatGPT dots, OpenAI's always-on agent, say it keeps stopping to ask for permissions and then forgets the approvals they already gave. The complaints began showing up on X on October 26 and were still coming in on October 27. Most come from people who say they pay for ChatGPT Pro or higher plans.

OpenAI launched dots last month as agents that run on their own cloud computer and keep working on tasks while you are away, as we covered at launch. The pitch was less babysitting. Several users now describe the opposite.

One user, posting as SpaghettiKind, wrote to OpenAI's Thibault Sottiaux that their dot "loses my approvals and simply stops working." The user said the dot had sent Gmail messages for them before, but its own tool now blocks it from doing so even after they approved it twice.

@thsottiaux We need to fix dots. My dot is growing increasingly useless. For example, it loses my approvals and simply stops working. It has gmail access, its sent emails for me before but now its “tool” is suddenly denying it from doing so after ive re-cleared it twice.

— SpaghettiKind (@MuhammadYa31086) October 26, 2026

A similar story came from developer Ziwei Guo, who said a long-running Codex task got stuck on a "Continue" button on an approval page and was still unfinished hours later, even with full access turned on. That post is about Codex rather than dots by name, but it describes the same kind of stall.

The most liked post in the group is a cancellation. The X user The Dude said they had cancelled their OpenAI subscription and called dots unable to do "the simplest task without breaking down and asking for 10 different bs permissions."

Just cancelled my @OpenAI sub.
@sama Dots is absolute trash. Can't do the simplest task without breaking down and asking for 10 different bs permissions.

— The Dude (@0xArvn) October 27, 2026

Usage limits are the other theme. AR, who says they pay $500 a month, wrote that their Codex meter fell from 93% to 33% in a morning with little use, then hit 0% after a single prompt. They said they have been burning through their weekly limit in a day. Another user, Rachel, said they have to use usage resets "all the time" and threatened to cancel their $200 and $500 accounts. Both replies were posted under a post by Sottiaux about GPT-6.1 Terra, which the post said gives twice the usage of GPT-6.1 Sol. A heavy user going by Andrew compared dots to Siri, calling it "over promised and highly disappointing."

OpenAI's status page showed all systems operational at the time of this writing, including ChatGPT Work and Codex. We could not find any public response from OpenAI or Sottiaux to these specific complaints.

It is worth keeping the scale in view. This is a small cluster of posts, and the biggest had a few dozen likes. These are individual user reports that we could not verify, and some of them mix dots complaints with broader frustration over model quality and Codex limits. We have not seen evidence of a wider outage.

Still, the complaints go to the core promise of dots. An agent meant to work on its own is not much use if it keeps waiting for a human to click approve.