OpenAI says Dots studies what it knows about you on its own. It still refuses read-only access to your drive

Dots, OpenAI’s assistant for subscribers paying $100 a month or more, regularly reviews what it knows about its user and looks for ways to help. Alexander Embiricos, who leads OpenAI’s enterprise product team, said so at TechCrunch Disrupt. He also mentioned talks with payment processors. For anyone weighing whether to let an agent into their inbox and calendar, that matters more than the feature list.

Dots is Codex in friendlier packaging, and it speaks up on its own

OpenAI released Dots at DevDay in late September. The assistant runs in the ChatGPT desktop and mobile apps for subscribers on the Pro tier or higher, with plans starting at $100 a month. Once connected to calendar, email, computer and other tools, it tries to help with work even when the user has not asked for anything.

Embiricos says much of that functionality was already possible in the Codex app, the coding tool released at the start of the year. It just took more work and hand-holding. Dots therefore rests mostly on framing: one entity and one conversation instead of many. The proactivity still has gaps. Early in development, his own assistant warned him about a calendar conflict three minutes before it started. TechCrunch’s testing confirmed that alerts arrive, though not always in time.

Asked whether Dots collects data the way Meta’s Muse does, Embiricos said yes. A feature called proactive research periodically goes over what the assistant knows about the user, their past conversations and what it has access to. Then it researches how it could help and proposes it. OpenAI says Dots has the same privacy controls as ChatGPT, including turning off model training on conversations. TechCrunch notes, however, that the controls are not granular and the app refuses read-only access to cloud drives.

Every moment of proactive thinking burns tokens

TechCrunch dryly points out that an assistant that needs no prompting means more token usage for OpenAI. The business model is built right into the proactivity. The more Dots thinks about you, the more compute runs and the more of your subscription you use.

The second layer is data. Meta’s Muse has drawn criticism for the profiles it builds of family and friends, which could also feed advertising. The startup Hark explicitly promises not to collect data for an advertising business, and TechCrunch puts OpenAI next to Meta on that point. Embiricos offers the model’s intelligence and discretion as the counterweight. For a company that would let Dots into work email, that is not enough. An IT admin needs to know exactly what the agent may do, and a reference to the model’s judgment does not tell them that.

Granular permissions exist only as a promise for now

Embiricos himself admitted OpenAI needs to build “super, super, super granular” controls. Dots does not have them today. Refusing read-only access sends a cautious user the opposite signal from the one they would expect: the agent wants more than it needs just to read something. After launch, OpenAI was actually criticized for the reverse problem, because Dots asked for permission too often. The company plans to loosen the defaults over time.

Scaling is still an open question too. Embiricos has a human assistant, because OpenAI’s corporate partners want to deal with a person. He says he never intends to have a meeting with someone else’s agent. The product chief selling the agent is confirming that the other side is not ready for agents yet.

Payment processors will decide whether Dots gets a wallet

Embiricos said OpenAI is deep in conversations with payment processors. He gave no details. If Dots could spend money, it would be a potentially lucrative first for OpenAI and the biggest concession users have made so far about what they hand to an agent.

It is worth watching whether OpenAI ships the promised fine-grained permissions, including read-only access, before payments. The second signal will be how proactive research appears in the privacy settings and whether it can be switched off on its own.

Lilith's verdict

Dots will not peek into the living room through the window, it asks for the whole flat, while OpenAI negotiates a wallet for it with payment companies. Whoever lets it in before getting locks for individual rooms will soon be reading statements for payments they never clicked.