ChatGPT Now Texts You, But Those Texts Are Not Fully Private

OpenAI has started letting some ChatGPT users text their "dot," the ChatGPT helper that works on tasks for them. The rollout began on October 28, 2026. The catch is privacy. A third-party company called Linq handles the texts on their way to OpenAI, so the conversation is not end-to-end encrypted.

What changed

Until now, a user worked with a dot inside ChatGPT. In this limited beta, an eligible user can hand the dot a task by iMessage or RCS and get updates in the phone's regular messaging app. RCS is the route on Android. The text thread reaches the same dot, with its existing memory and connected plugins.

Texts can include words, photos and voice messages. The dot can also message first when it has something worth reporting. If it needs approval for an action, it sends a link, and the review happens in ChatGPT, not in the text thread.

Who gets it

• The beta is limited to US Pro users, and a Pro plan alone does not guarantee access.
• Business and Enterprise workspaces are not included.
• SMS and group texts are not supported.
• Users connect a phone in the dot's Texting setup and verify the number.

The chat is one-to-one. The account holder texts their own dot, and the dot texts back. Coworkers or clients cannot be added. Texting also does not give the dot or OpenAI access to the user's other phone conversations.

The privacy catch

Linq processes messages on the way to OpenAI. So the iMessage or RCS label on a phone does not mean the whole conversation is end-to-end encrypted. End-to-end encryption means only the sender and receiver can read a message.

That matters for anything sensitive. An unpublished transcript, a confidential brief, customer details or a password pasted into a text would all travel through that route.

What to do

The safest habit is to send a short instruction, not the material behind it. For example, a creator could ask the dot to summarize an interview already stored in a connected project, instead of pasting the transcript into the thread. That only works if the creator is allowed to use that project with the dot.

Sending a text does not connect a new inbox or project. Still, leaving a sensitive passage out of a text does not settle whether the dot can already read it through an existing connection. Those permissions are worth checking too.