News · 2026-10-29

OpenAI brings Dot texting to a limited US Pro beta

OpenAI added texting to Dots on October 28, allowing eligible United States Pro users to reach the same persistent agent through iMessage or Rich Communication Services messaging. The limited beta carries existing memory and connected-app context into a phone conversation and can deliver proactive updates. It expands access to an ongoing agent rather than launching Dots for the first time.

Key facts

• The texting beta is restricted to eligible US Pro accounts and may not reach every Pro user immediately.
• Supported routes are iMessage and RCS; OpenAI does not list individual carriers.
• Dots launched on September 29, almost a month before the texting update.
• Primary source: OpenAI’s October 28 release notes.

The distinction between the original product and the new channel matters. OpenAI’s Dots introduction describes an always-on agent with its own cloud computer, connected applications, memory, and ongoing work. Texting makes that continuing entity easier to reach while away from the application. It does not create an unrelated assistant that starts with an empty history each time a message arrives.

A useful analogy is calling the same assistant at a different extension. The telephone changes the route into the conversation, but the assistant still has the same notes and responsibilities. That continuity is the benefit: a user can check progress or give a new instruction without reconstructing the task. It is also the privacy issue, because information accumulated through one channel can inform work requested through another.

OpenAI’s setup guide narrows the availability. Texting is not available in Business or Enterprise workspaces, and access remains a limited beta. The guide identifies iMessage on iPhone and RCS on Android as the supported routes. Describing this as universal text-message access would obscure both the account restrictions and the protocol requirements.

The release notes identify Linq as the third-party messaging provider. OpenAI says ordinary message and data charges may apply, and users can reply STOP to stop outgoing texts. It also says the Dot can text only its owner, only its owner can text it, and the Dot cannot read the owner’s other text conversations. Those are published product boundaries, rather than findings from an independent security test.

The more consequential control issue concerns retained context. OpenAI’s current help materials say users cannot inspect, edit, or delete individual Dot memories. Deleting the Dot removes its own context, but does not delete separately stored files, Codex threads, or ChatGPT conversations it created. Disconnecting a plugin prevents new access without erasing information already incorporated into that context. Our lesson on agent memory explains why storage, retrieval, and deletion should be treated as separate operations.

The privacy and safety FAQ also separates proactive research from actions. Proactive research can read permitted sources and save private notes, but it cannot itself send messages, change content through plugins, or control a browser or computer. Follow-up actions remain subject to the product’s action rules and safety checks. A proactive text update therefore should not be treated as evidence that a Dot has unrestricted authority to act through connected accounts.

For personal plans, OpenAI identifies the “Improve the model for everyone” setting as a control over whether conversations and work may improve models. It says it does not train directly on proactive background research or private notes, while information later used in an eligible conversation or task may be treated according to the user’s settings. These distinctions concern OpenAI’s stated policies, not a claim that every possible downstream data use has been independently audited.

The Reddit discussion captured both attraction and caution. Some commenters worried about impersonation, connected-account permissions, and prompt injection. Others emphasized that texting is a new communication channel, rather than permission to impersonate the owner. The thread demonstrates concerns, not a confirmed exploit or a representative survey.

This distinction matters when stopping work as well. Stopping outgoing texts is a messaging control; deleting retained context is a separate operation with separately documented consequences. A user should not infer that silencing a channel erases the persistent agent’s information or all of the artifacts it has already created.

The practical caveat is that convenience does not settle control. A persistent agent reachable from a pocket is useful precisely because it remembers and keeps working. Readers should evaluate the beta with its actual account scope, messaging provider, memory limitations, and temporary introductory usage terms attached. The new feature is reachability; the continuing questions are what the agent remembers and what it is allowed to do.

Primary source, verified: read the paper →

Ground Truth. (2026, October 29). OpenAI brings Dot texting to a limited US Pro beta. Ground Truth. https://groundtruth.day/news/openai-dots-texting-us-pro-beta.html

@misc{groundtruth:openai-dots-texting-us-pro-beta,
  title  = {OpenAI brings Dot texting to a limited US Pro beta},
  author = {{Ground Truth}},
  year   = {2026},
  month  = {oct},
  url    = {https://groundtruth.day/news/openai-dots-texting-us-pro-beta.html}
}

Topics: openai · agents · messaging · privacy · product-releases