OpenAI Unveils Dots Agent and Promises a 'New Privacy Standard' in AI

At the OpenAI DevDay conference, Sam Altman introduced Dots, a new AI agent, and said the company wants to "set a new standard for privacy in frontier AI." All day, OpenAI took jabs at its main rival, Meta's Muse agent, criticizing it for poorly protecting user data. The irony is that Muse launched only a couple of months ago as a safer replacement for OpenClaw. Mark Zuckerberg promised at the time that the product was "built from the ground up for privacy and security." The two largest labs are saying almost the same thing, and users are left to decide whom to believe.

One Promise for Two

The market logic is clear. Companies accumulate personal data, leaks happen constantly, and trust in AI products is fragile. Privacy has moved from a legal clause to a marketing argument: it is put on stage, placed in the headline of an announcement, and used to hit a competitor.

The problem is that words on stage are harder to verify than technical specifications. A claim of a "new standard" does not say what exactly has changed: where data is stored, for how long, who has access to it, and whether it is used to train models. So far, the public materials show only rhetoric. Muse has already been through the cycle of a loud promise followed by questions about implementation, and now Dots is heading down the same path.

Why Agents Have It Harder Than Chatbots

A chatbot knows what you typed into it. An agent acts on your behalf: it needs access to email, calendar, files, the browser, and sometimes payment tools. The volume of data and the cost of a mistake both grow by an order of magnitude. This is a general feature of the category rather than a detail of any specific product, but it is exactly what makes privacy promises so weighty.

Agents also bring risks of their own. These include instructions embedded in external web pages that nudge the model toward unwanted actions, long-term memory containing personal information, and detailed operation logs. Each of these can become a source of leaks, even if the stated policy is flawless.

What We Know So Far

• Dots: a new OpenAI agent, presented at DevDay by Sam Altman himself.
• Stated goal: a "new standard for privacy in frontier AI."
• Muse: Meta's agent, launched roughly a couple of months before DevDay and positioned as a safe successor to OpenClaw.
• Meta's promise: the product is "built from the ground up for privacy and security" (Zuckerberg's words).
• OpenAI's stance at DevDay: indirect criticism of Meta for handling data unsafely.
• Not disclosed in the available material: technical guarantees, data retention periods, and an independent audit of either product.

How to Treat These Promises in Practice

For an ordinary user, and even more so for a company, the rule is the same: look at the settings and documentation, not the slogan. Before connecting an agent to work systems, it is worth finding out several things:

• Whether your data is used to train models by default, and whether you can turn this off.
• How long the action history and the agent's memory are stored, and whether you can delete them.
• Whether permissions can be granted narrowly, for example read-only access to the calendar rather than the whole mailbox.
• Whether there is an agent action log and a confirmation step for risky operations such as payments and sending emails.
• Whether the vendor has published the results of an independent security audit.

Until these questions have answers, it makes sense to start with a limited pilot on non-critical data, keep the agent away from payment and corporate systems, and recheck the claims after several months of use. Muse has already been through this, and now Dots faces the same test.