OpenAI Hands ChatGPT the Keys to iMessage on Mac

OpenAI just gave its chatbot direct access to one of the most personal corners of the Apple universe. On Thursday, the company rolled out an Apple Messages plugin for ChatGPT on Mac. The tool lets the AI read conversations, summarize threads, draft replies and even hit send. Users of ChatGPT Work and Codex plans can activate it now.

But the move lands amid a complicated relationship between the two companies. Apple and OpenAI struck a partnership in June 2024 to bring ChatGPT into Siri and Writing Tools across iOS 18, iPadOS 18 and macOS Sequoia. Months later Apple sued OpenAI, accusing the startup of trade secret theft. The lawsuit remains active. Bloomberg reported the new plugin Thursday and highlighted the potential for fresh privacy friction.

The feature requires explicit user consent at every step. On a Mac running Apple silicon, owners must grant Full Disk Access in System Settings. They also approve contact automation permissions. Once enabled, ChatGPT operates locally. OpenAI states the model does not store message content or build a full index of conversations on its servers. The company repeated that assurance in its announcement.

Still. Granting any AI full read and write access to years of text messages carries risks. iMessage holds family plans, work negotiations, health updates and late-night confessions. A single misconfigured permission or future software bug could expose that data. Security researchers and privacy advocates have already begun to question whether local execution truly walls off every vector.

OpenAI demonstrated the plugin in a short video. In the clip, a user asks ChatGPT to review recent messages from a book club thread. The AI surfaces key details about an upcoming meeting, suggests a polite reply and offers to send it. The process takes seconds. Early testers on X described the workflow as surprisingly fluid. One post noted that people already paste iMessage screenshots into ChatGPT; this simply removes the copy-and-paste step.

The timing feels deliberate. Apple has spent years marketing itself as the privacy-first platform. Its on-device Apple Intelligence features process many tasks without sending data to the cloud. Yet the company itself chose to integrate ChatGPT for more complex queries, with user approval required each time. Now OpenAI is reaching deeper into the same device with broader ongoing access. The contrast raises an obvious question. Who actually controls the data once permissions are granted?

Analysts see this as part of a larger pattern. OpenAI continues to push agent-like capabilities. Its newer models can run for extended periods, use tools and take actions across apps. The Messages plugin fits that strategy. It turns a conversational chatbot into something closer to a personal assistant that can act inside the communication layer millions rely on daily. And it does so on the desktop first, where power users and enterprise customers experiment most freely.

Apple has not commented publicly on the plugin. Its own integration with ChatGPT remains more limited and transactional. Siri can hand off a query. Writing Tools can call on the model to rewrite text. Both interactions require fresh consent and do not grant persistent access to the Messages database. The new OpenAI feature crosses that line. Whether Apple views it as competitive overreach or simply another third-party extension remains unclear.

Privacy experts point to past incidents. OpenAI has faced multiple lawsuits over data handling, training practices and unauthorized use of personal information. The company has tightened safety processes in recent weeks following other security incidents. Yet each expansion of capability invites new scrutiny. Full Disk Access on macOS is a powerful entitlement. It opens Mail, Safari history, backups and more. Users must weigh convenience against that breadth.

Early reaction on X split along predictable lines. Some called the feature inevitable and useful. Others advised turning it off entirely. One user wrote that AI searching, organizing or sending messages now exists just like camera phones once did. The choice belongs to the individual. Another warned that full disk access reaches far beyond Messages alone.

The plugin arrives as Apple prepares further AI updates. Its own models have improved in on-device performance. Yet the decision to partner with OpenAI signaled that even Apple needed outside help for certain tasks. That dependency creates an opening. OpenAI is using it.

For now the rollout targets Mac users on paid plans. Mobile integration does not exist. The company says the feature runs only on-device and respects user controls. Those statements matter. They also invite testing. Security teams will likely probe how strictly the sandbox holds once the model begins to act on message content.

The broader contest between Apple and the AI leaders keeps intensifying. Google, Meta and others race to embed intelligence into communication apps. OpenAI’s move stands out because it targets the closed messaging garden Apple has protected for years. iMessage remains a key reason many stay inside the Apple ecosystem. If users grow comfortable letting ChatGPT manage those conversations, the stickiness could shift.

Practical uses seem obvious. Busy professionals might ask the AI to triage group chats, flag action items or suggest responses during meetings. Parents could request summaries of school threads. The productivity gains feel real. The privacy trade-off feels personal. Each user will calculate that equation differently.

OpenAI has positioned the plugin as an opt-in tool, not a default. It requires the @ mention or menu selection to activate. The company stresses that no messages leave the Mac. Those details aim to calm concerns. Whether they succeed will show in adoption rates over the coming weeks.

The episode also highlights how quickly capabilities evolve. What began as a cautious partnership in 2024 has become a public contest over access and control in 2026. Apple wants to keep sensitive data close. OpenAI wants to demonstrate that its models can act responsibly with that data when permitted. The tension shows no sign of easing.

Industry watchers will track two metrics. First, how many Mac users enable the plugin despite the permissions required. Second, whether Apple responds with tighter controls, new warnings or its own expanded messaging intelligence. The answers will shape the next chapter of AI inside the devices people trust most with their private lives.

One thing is already clear. The barrier between conversation and computation just lowered again. ChatGPT can now participate in iMessage not merely as an advisor but as an active participant. The implications stretch beyond a single plugin. They touch on trust, control and the future shape of personal communication.