OpenAI’s new Apple Messages plugin moves ChatGPT from drafting text in a separate window to acting inside a user’s message history. On supported Apple silicon Mac computers, users can ask ChatGPT Work or Codex to search and summarize conversations, locate information, draft replies, and send messages through Apple’s Messages app across iMessage, SMS, and RCS—the modern successor to SMS. The plugin requires macOS permissions, and OpenAI says it reads messages in response to specific requests, processes and stores message content locally by default, and requires approval of the message and recipients before sending on a per message level—unless the user grants persistent permission.

Lawyers should consider how this new feature impacts the confidentiality and privilege of messages they send their clients via Apple's Messages app. Local processing is reassuring but not dispositive. In an area of unsettled law, the critical questions to be aware of are: (1) whether the plugin processed communications between an attorney and client; (2) whether that processing exposed content beyond the client's device; and (3) whether the client acted consistently with an expectation of confidentiality.

Lawyers are not going to be able to control all of their client's devices, but we should be having conversations with our clients about communicating through devices or accounts to which third parties (through AI or otherwise) have access. And for lawyer devices, law firms should treat message-enabled AI as a governed technology rather than a personal convenience. As this develops, sound practices include keeping per-message approval enabled, using only firm devices to communicate with clients, and avoiding the use of text messages for highly sensitive matters.