OpenAI has introduced a Mac integration that lets ChatGPT read, search and act on conversations in Apple’s Messages app with user permission. While the company says the feature operates locally and requires several opt-ins, the move could raise questions about access to sensitive personal communications.
ChatGPT users on Mac can now give the AI chatbot access to Apple’s Messages app, allowing it to search conversations, summarise chats and, in some cases, read, write and send text messages.
The new capability, reported by Bloomberg’s Mark Gurman, expands OpenAI’s growing set of app integrations on macOS. But Messages is a more sensitive addition than tools such as Notes, Terminal or Xcode, as it could give an AI system access to conversations that users may consider among the most private data stored on their devices.
OpenAI says the integration requires explicit permission and relies on tools already available within macOS. The company has also sought to address concerns that the feature could result in a blanket collection of users’ message histories.
The development nevertheless places OpenAI’s ambitions alongside Apple’s long-standing effort to position privacy as a central part of its product strategy.
How ChatGPT’s Messages integration works on Mac
The integration is not switched on by default. Users who want ChatGPT to interact with Messages must go through several permission settings within both the ChatGPT app and macOS.
During setup, ChatGPT displays a notice explaining that the application will be able to access message history stored on the Mac. Users must also adjust permissions in System Settings, including enabling Full Disk Access.
More from Tech
Additional access is required for information such as contact names and for the automation mechanisms that allow ChatGPT to perform actions within the Messages app.
According to OpenAI, the integration runs locally on the Mac and uses existing macOS features, including AppleScript and Accessibility, to communicate with Messages. The company said it does not automatically build an index containing all of a user’s conversations.
Those safeguards are significant because access to Messages potentially exposes far more than simple text exchanges. Conversations can contain personal information, financial details, addresses, work discussions and other sensitive material. Even when access is voluntary, the feature is likely to prompt closer scrutiny over how much data an AI assistant can process and under what circumstances information may leave a user’s device.
Apple has already placed similar restrictions around the use of ChatGPT elsewhere in its ecosystem. OpenAI recently introduced support for analysing information from the Health app on iPhone and iPad, but Apple’s operating systems require users to explicitly approve the sharing of that data.
A new point of tension between Apple and OpenAI
The Messages integration also arrives as the relationship between Apple and OpenAI faces broader pressure.
Apple sued OpenAI last month, accusing the AI company of obtaining information related to unreleased products. Against that backdrop, an OpenAI feature capable of interacting with one of Apple’s most important communication services could add another point of friction between the companies.
The move may also overlap with capabilities Apple is developing for its own AI-powered version of Siri. Apple has been working on features that would allow Siri to understand and retrieve information from personal content, including messages stored on a user’s devices. Such access could become an important distinction between Apple’s assistant and third-party AI services.
Apple has previously taken action against products that attempted to extend or alter access to iMessage. Beeper Mini, for example, was shut down after offering a way for Android users to communicate through Apple’s messaging service.
For now, ChatGPT’s Messages support remains dependent on users deliberately enabling the required permissions. That makes it different from unrestricted access to an iPhone’s communications.
Still, the integration highlights a broader question that is becoming harder to avoid as AI assistants gain control over personal software: users may increasingly have to decide how much access they are willing to grant in exchange for automation and convenience.