Texting your Dot: iMessage, RCS and privacy

In this article

An update on ongoing work does not always arrive while someone has ChatGPT open. On October 28, OpenAI announced a messaging channel for talking to a personal Dot and receiving proactive updates. Called “Texting your dot,” it began rolling out in beta to eligible Pro users in the United States.

The feature uses iMessage or RCS and shares the Dot’s memory and plugins with its ChatGPT experience. It is another way to reach an already configured agent. It does not give every ordinary ChatGPT conversation a phone number or establish that texting is available in Brazil.

The agent and its messaging channel do different jobs

OpenAI describes a Dot as a persistent agent that can follow goals and continue work between conversations within the conditions set by its user. A regular conversation responds to requests; an ongoing agent’s context may include tasks, connected tools and commitments to follow up.

Messaging provides a route to that same Dot outside the usual interface. It does not independently create new permissions over other accounts. If a task needs an app, the app’s connection and access still matter. Receiving a reply on a phone is also not evidence that every part of a project is finished.

The current Dot documentation separates eligibility for the agent from eligibility for texting. That distinction matters when deciding whether a general product announcement applies to a particular way of using it.

The beta has its own boundaries

The October 28 release notes name Pro users in the US. The guide checked on November 4 continues to describe texting as a beta for a limited number of Pro users there and says the channel is not supported in Business or Enterprise workspaces.

The broader Dot product has other access conditions, but those should not be transferred to texting. Buying Pro outside the US is not a guarantee of receiving this beta. The consulted sources do not establish a texting rollout date for Brazil.

For eligible accounts, the guide directs users to the Dot’s profile, then Texting and the official number-verification flow. iMessage must be enabled on an iPhone, or RCS on an Android phone. A missing option may reflect eligibility or gradual rollout. IANautaLab did not activate the beta or register a phone number for this report.

The practical starting point is therefore the account’s actual interface. A screenshot from someone else’s eligible account cannot establish access for a reader in a different region or workspace.

What useful follow-up could look like

A hypothetical example is asking about a presentation the same Dot is already preparing. Another is receiving an update about a task it has been asked to follow. Shared context can help continue the work without restating every detail in each message.

These examples depend on a genuinely configured task and available capabilities. An update may identify something that needs review instead of presenting a final decision. Important output should still be checked, including what has been completed and what remains outstanding.

A short message can also be ambiguous. “Go ahead” does not, on its own, define which external action is authorized. Specifying the goal, the limits of independent action and the decisions that must come back to the user makes progress easier to assess, whichever app carries the conversation.

This relates to IANautaLab’s explanation of proactive agents and access limits, although the products are different. Receiving an update before asking another question does not remove the need to control the sources an agent consults or the actions it may take.

Where Linq enters the conversation

OpenAI uses Linq as a messaging intermediary. Its current guide explicitly says the user-to-Dot conversation is not end-to-end encrypted along this route, including when iMessage or RCS is used. The familiar messaging interface should not be interpreted as a guarantee that only the two endpoints can process the content.

According to OpenAI, Linq processes texts for delivery without retaining their content afterward, though limited metadata may be retained. Texts also become part of the Dot conversations stored in ChatGPT. These are documented vendor statements, rather than findings from an IANautaLab security audit.

The documentation says enabling texting does not grant the Dot or OpenAI access to other text conversations on the phone. That limits the integration’s scope, while the messages deliberately sent to the Dot still pass through the documented services. Sensitive information calls for evaluating that route before sending it.

A familiar conversation still has technical limits

The guide describes one-to-one messaging between an owner and their Dot. It cannot participate in group texts. SMS is not supported, and the Dot’s texting number is not a destination for telephone calls.

A phone’s general ability to send a text is therefore insufficient. The enabled protocol and eligible account form part of the feature. It also should not be assumed that a Dot can start messaging anyone in the user’s contacts.

Carrier message or data charges may apply. The ChatGPT plan, beta eligibility and telecommunications charges are separate considerations. A subscription should not be presented as covering every possible cost, or recommended as a way to guarantee regional access that has not been announced.

These constraints are worth checking before making texting part of an important workflow. An update channel that is useful for routine follow-up is not automatically a dependable substitute for every other way of reviewing work.

Turning off texting does not erase the record

The documentation offers STOP or turning off Texting in the Dot’s profile to stop messages. That disconnects the channel and number, but does not automatically delete previous conversations or memories. Managing stored information is a separate decision.

Eligible users can follow work through an everyday communication channel. Readers outside the beta do not yet have a guaranteed activation process in the consulted documentation. In either case, useful oversight includes knowing what the agent did, which information it used and where review is still needed.