ChatGPT's Mac app can now work directly with conversations in Apple's Messages app. OpenAI added an optional Apple Messages plug-in on August 20 that can read and search chats on a Mac, prepare replies, and send messages after user approval under the default settings.
The integration is available across ChatGPT plans in the macOS desktop app and can also be used in ChatGPT Work and Codex. OpenAI's release notes make the control boundary clear: sending is not supposed to happen silently by default. The user approves both the message and its recipients before it goes out.
It works with the Messages data already on your Mac
OpenAI says the plug-in can work with iMessage, SMS and RCS conversations available through Messages on the Mac. MacRumors reports that the feature is limited to Apple-silicon Macs and requires the macOS permissions needed for ChatGPT to access local message data and automate the Messages app.
That distinction matters. This is not a new ChatGPT chat channel that lets somebody remotely contact the assistant through iMessage. It is a local Mac integration that lets an enabled ChatGPT workflow act on conversations already available to the user's Messages app.
Practical uses include searching for a detail buried in a conversation, summarising a thread, drafting a response or preparing a message based on information from another task. The most consequential step is still sending, which is why approval settings deserve attention.
Persistent approval trades friction for risk
OpenAI specifically warns about persistent approval. Users can reduce repeated confirmation prompts in some workflows, but doing that also reduces the point at which a person can catch the wrong recipient, an awkward draft or a task that has interpreted instructions too broadly.
There is also a documented edge case for tasks that disable approval prompts: Messages may be unable to display the confirmation required to send. OpenAI recommends switching back to an approval mode when that happens.
For most people, the safer starting point is the boring one: enable the plug-in only if the workflow is useful, keep send approval on, and review the message and recipient before confirming.
This is a Mac feature, not an iPhone replacement for Messages
The current release is about the ChatGPT desktop experience on macOS. It does not turn the iPhone Messages app into a ChatGPT client, and users should not assume that a Mac integration automatically carries across Apple's other devices.
The more interesting shift is what happens on the desktop. ChatGPT can increasingly combine information across apps and then take an action in the app where the work belongs. Messages is a particularly sensitive example because the output goes to another person, not just into a private note or search result.
That makes the approval model as important as the capability itself. Reading and drafting can save time; sending changes the real world. OpenAI's defaults recognise that difference, and users should too.
The permissions are worth reviewing before the first useful prompt, not after an accidental overreach. MacRumors notes that the integration needs access such as Full Disk Access, contacts and automation capabilities to do its job. Those permissions are powerful by design. If the plug-in stops being useful, disabling it and revoking permissions is a cleaner choice than leaving broad access enabled indefinitely. Convenience is strongest when the user still understands which app can read which data and which actions require confirmation.
Reporting notes