ChatGPT’s Textable Dot Adds a Messaging Route to an Always-On Agent

An always-on assistant is only useful if it is available when a person needs it. OpenAI’s new texting beta brings its “dot” assistant to iMessage or RCS, alongside its ChatGPT and team-messaging surfaces. The convenience is obvious: a user can send a quick instruction from a phone. The less obvious point is that this is not an isolated SMS bot. OpenAI says the dot shares memory and connected-app context with ChatGPT, and Linq handles the messaging connection.

What the beta does

OpenAI’s October 28 release note says texting is starting to roll out in beta to Pro users in the U.S. The setup uses iMessage on iPhone or RCS on Android, and the company says a dot can text proactively with updates. OpenAI says the dot can only text its user and that this connection does not give OpenAI access to other text conversations. The company identifies Linq as its third-party provider for sending and receiving texts. ^[1]

The current Help Center narrows eligibility further: access is limited, not available to every Pro account and not offered in Business or Enterprise workspaces. It advises caution with sensitive information, notes possible message and data charges, and provides “STOP” as a way to stop outgoing messages. These are beta terms and can change, so the live setup screen and linked notices should be checked before enabling the feature. ^[2]

The dot has more context than a standalone text bot. OpenAI’s product description says dots can work across connected apps and carry context across channels. The Help Center also says connected apps may be reviewed proactively, disconnecting an app does not delete information already obtained, and some actions can require approval or handoff. The consequences of a text command therefore depend on the dot’s broader access and rules, not just the phone number receiving the message. ^{[3] [2]}

A new channel, same authority questions

Messaging makes interaction quick, but the channel can blur distinctions between low-risk questions and consequential instructions. A short text such as “send the update” may be ambiguous about the recipient, content or timing. If the assistant has access to email or work apps, the request could be interpreted using that connected context. A good design asks for confirmation when the action or target is unclear and makes the pending action visible before it happens.

OpenAI describes custom rules that can allow an action, require a specific instruction, ask before proceeding or hand the decision back to the user. Those settings are useful only if users understand their scope and can review them. The help article warns that asking for a draft does not authorize sending it, and says connected apps can have different read and write permissions. A text channel should preserve those distinctions rather than turn brevity into blanket consent. ^[2]

There is also a third-party boundary. OpenAI names Linq as the messaging provider, but public announcement material does not establish all details of message retention, processing locations, support access or how each failure is handled. Users should read the notice shown during setup and the current privacy terms instead of inferring the entire data path from the fact that a protocol is iMessage or RCS. End-to-end encryption properties of a messaging service do not by themselves answer what a connected service provider receives when a message is routed to an assistant.

Practical checks before connecting

Before enabling a text connection, check which account and dot are being linked, what memories and apps that dot can use, what tasks are scheduled, and which actions are allowed without approval. Remove access that is unnecessary. Avoid sending passwords, payment information, health records, customer secrets or sensitive work details through a beta messaging path unless the relevant terms and organizational policy permit it.

Try a low-stakes request first. Confirm where the answer appears, whether the dot can initiate messages, how a task is paused, and what “STOP” actually stops. The Help Center says replying STOP halts outgoing texts, while resetting a dot is the mechanism described for deleting its conversations and saved memories. Those are different controls; one stops the channel, the other removes the assistant state. ^[2]

For a workplace, the current release notes say texting is not available in Business or Enterprise workspaces. More broadly, organizations should evaluate third-party processors and assistant permissions before staff route sensitive information through personal messaging accounts. The FTC’s data-protection guide emphasizes knowing what personal information is collected, retaining only what is needed and limiting access; it is general guidance rather than a review of OpenAI, Linq or this beta. ^[4]

Convenience is a permission surface

The change is small in interface terms: a dot becomes reachable by text. In product terms, it brings an always-on agent, its memories and connected tools into a channel designed for quick, informal messages. That can make the assistant more useful for reminders and lightweight follow-ups. It also means the user should know which context follows the conversation and what actions a short instruction could trigger.

OpenAI’s stated beta boundary—eligible U.S. Pro users, gradual rollout, third-party provider, caution with sensitive data—is specific and worth keeping in view. The feature does not imply access to the user’s other text conversations, according to the company. The prudent approach is still to review the connection notice and permissions, keep consequential actions behind explicit approval, and treat the text route as another entrance to the same assistant authority.