AI Takes Over Texting as ChatGPT Gains Access to Apple Messages

OpenAI has given ChatGPT a new role on the Mac: helping users manage their conversations inside Apple's Messages app, including searching message histories, drafting replies and sending texts.

The new Apple Messages plugin, launched on August 20, allows ChatGPT to work with iMessage, SMS and RCS conversations stored in Apple's Messages application on compatible Macs. The integration is available through the ChatGPT desktop experience and is currently limited to Apple Silicon Macs.

The change moves ChatGPT beyond simply helping people write a message. Users can ask the AI to search through conversations, analyse exchanges, suggest follow-ups and prepare responses. With permission, it can also send messages through Apple's Messages app, effectively turning the chatbot into an assistant capable of taking action on a user's behalf.

OpenAI's suggested use cases illustrate how broad that role could become. A user could ask ChatGPT to find birthdays mentioned in messages and add them to a calendar, identify potential spam, or look through recent conversations for follow-ups that need attention. The idea is to turn a previously manual task, searching through messages and deciding what to do next — into something an AI agent can handle conversationally.

But there is an important limitation: this is not ChatGPT taking over texting on the iPhone. The new integration operates through the Mac's Messages environment. It requires users to grant system-level permissions, including Full Disk Access, access to contact names and automation capabilities. Sending messages requires user approval by default, and OpenAI advises against giving the system persistent approval.

That permission model is likely to become one of the most closely watched aspects of the rollout. Messages can contain highly personal conversations, financial information, photographs, passwords, business discussions and details about people who never agreed to have an AI system analyse their communications. Giving an AI access to that archive therefore raises a different category of privacy question from asking it to write a standalone email or social-media post.

The timing is also notable because Apple and OpenAI are already locked in a legal dispute. Apple sued OpenAI in July, alleging that the company misappropriated trade secrets, while OpenAI has denied the allegations. The new Messages capability therefore arrives as the two companies are simultaneously partners in Apple's broader ChatGPT integration and adversaries in court.

Apple already incorporates ChatGPT into parts of Apple Intelligence, including Siri and system-wide Writing Tools. Apple says users can call on ChatGPT for tasks such as answering questions, analysing documents and images, and helping compose content. The newer Messages plugin is different because it gives ChatGPT a direct operational pathway into conversations already stored on a Mac.

That distinction could become increasingly important as the AI industry moves from chatbots that generate answers to agents that perform tasks. Reading a message, deciding what it means, composing a response and sending it are separate steps that previously required a human to move between applications. ChatGPT is now attempting to connect those steps into one workflow.

The development is already prompting a broader debate about authenticity. If an AI writes and sends a personal message, the recipient may have no immediate way of knowing whether the words came from the person they are communicating with or from an AI assistant acting on that person's instructions. That may be convenient for routine communication, but it becomes more complicated when the conversation involves relationships, apologies, negotiations or emotionally sensitive matters.

For Apple, meanwhile, the move highlights the growing contest over who controls the user's personal digital context. Apple has its own AI ambitions, including Smart Reply and message summaries within Apple Intelligence, while OpenAI is increasingly positioning ChatGPT as an assistant that can move across applications and perform actions rather than simply answer questions.

The broader significance is that AI is moving another step closer to becoming an active participant in everyday communication. The question is no longer simply whether machines can write convincing messages; it is whether people are comfortable allowing them to read conversations, decide what requires a response and send that response. As these systems become more capable, convenience will increasingly have to be weighed against privacy, consent and the very human expectation that the person on the other end of a conversation is actually the one doing the talking.