OpenAI has added a new plugin to ChatGPT’s Mac app that lets it read, search, and send messages through Apple’s Messages app — covering iMessage, SMS, and RCS conversations — with each outgoing message requiring manual approval before it sends.
That approval step is the detail worth focusing on. Rather than letting the assistant fire off replies autonomously, OpenAI built in a confirmation gate, which suggests the company is being deliberate about liability and user trust around an integration that touches personal conversations directly. It’s a meaningfully more sensitive category of access than, say, letting an AI assistant read calendar events or draft an email in a separate window.
The feature slots into a broader trend of AI assistants requesting deeper OS-level access on Mac — Apple’s own Apple Intelligence has been inching toward similar territory, and Meta’s newly launched AI Mac app includes window-sharing and dictation tools that push in a similar direction. The competitive logic is straightforward: whichever assistant can actually act inside a user’s existing apps, rather than just chat in its own window, has a stickier reason to stay installed.
What’s not yet clear is how this handles group chats, message history retention, or whether conversation content is used in any way beyond the immediate session — OpenAI hasn’t published a detailed technical breakdown of data handling for this specific plugin. Anyone with sensitive conversation threads should look at the permission settings before granting access.Expect scrutiny here. Messaging-app integrations for AI assistants tend to draw attention from privacy researchers quickly, and this one touches a genuinely private data category. How OpenAI responds to that scrutiny, more than the feature itself, will likely shape whether this becomes a standard capability across other AI assistants on Mac.
