OpenAI's Apple Messages Plug-In Lets ChatGPT Text For You

OpenAI has launched an Apple Messages plug-in for ChatGPT that can search, draft, send, and delete texts on a user's behalf.

The plug-in also works with Codex and ChatGPT Work, so it reaches work accounts, not just personal ones. Users connect their Messages inbox to the chatbot, then sort, analyze, or edit messages directly from ChatGPT.

What the plug-in can do

• Sort, analyze, or edit messages from inside ChatGPT
• Draft and send messages on your behalf
• Delete messages for you
• Search for information buried deep in your message history

A brief commercial advertising the plug-in shows a user asking the chatbot to suggest follow-up messages to their contacts based on the messages received the previous day — the clearest picture so far of the workflow being sold.

What OpenAI says about the privacy question

OpenAI told Bloomberg that the plug-in runs locally on a user's machine and that it “doesn't create an index of all someone's messages.” TechCrunch, which reported the launch, wrote that the specifics of what that means aren't immediately clear, and said it had asked OpenAI for more information.

Read that as the company's own characterization, not a verified architecture. “Runs locally” and “no index” are two separate promises, and neither one says which message text leaves the device when you ask a question that needs a model to answer it. Until OpenAI describes the data path, the honest position is that the scope is undocumented.

The setting to leave switched off

On message sending, OpenAI encourages users to keep an eye on what ChatGPT is doing and discourages turning on persistent approval, warning that doing so “removes your final chance to review a message before ChatGPT sends it as you.”

That is a vendor warning people off its own convenience toggle, which is worth taking at face value. Per-message approval is the difference between an assistant that proposes text and one that speaks in your name to your contacts.

Is it worth connecting?

The case for it is retrieval: message history is one of the few personal archives with no decent search, and a chatbot that can find the address someone texted you in March earns its keep. Drafting is a weaker case, because the review step you should keep costs about as much attention as writing the reply yourself.

The case against is that a messages inbox is not only your data. Group threads and one-to-one conversations contain other people's words, sent to you under an assumption about who would read them. Nobody in those threads consented to the plug-in. If you work under confidentiality rules — legal, medical, HR, security — that alone is a reason to leave it unconnected until your employer has a position on it.

What to watch

Two things: whether OpenAI answers the outstanding questions about what “runs locally” actually covers, and whether ChatGPT Work gives admins a way to allow or block the plug-in centrally. The second one decides whether this stays a personal choice or becomes an IT policy item.