ChatGPT Can Now Read, Send, and Delete Your Apple Messages

Key points

• OpenAI launched an Apple Messages plugin for ChatGPT on 20 August 2026, giving the chatbot access to a user's iMessage inbox.
• The plugin can draft, send, delete, and search messages, as well as suggest follow-up replies based on previous conversations.
• OpenAI says the plugin runs locally on the user's device and does not build a searchable index of all messages.
• OpenAI explicitly warns users not to enable persistent approval, because doing so removes the final review step before ChatGPT sends a message in your name.
• The plugin also connects with Codex and ChatGPT Work, OpenAI's workplace tools, so it is not limited to personal use.

Imagine you wake up Monday morning to seventeen unread messages. Instead of scrolling through them yourself, you ask ChatGPT to summarise who needs a reply and suggest what to say. That is exactly what OpenAI's new Apple Messages plugin is designed to do.

The plugin, launched on 20 August 2026, connects your iMessage inbox directly to ChatGPT. From there, you can ask the chatbot to sort conversations, draft replies, send messages on your behalf, or dig up something a friend texted you six months ago that you definitely cannot find yourself.

What can it actually do?

Quite a lot. A short promotional video OpenAI released shows a user asking ChatGPT to suggest follow-up messages based on the previous day's conversations. You can also instruct it to delete messages or send a reply without you typing a word.

The plugin ties into Codex and ChatGPT Work, which are OpenAI's tools aimed at professional settings, so teams could theoretically use it to manage work communications too.

Should users be worried about privacy?

That depends on your comfort level, and it is worth being clear-eyed about the trade-off.

OpenAI told Bloomberg that the plugin runs locally on your machine, meaning the processing happens on your own device rather than on a remote server. The company also said it "doesn't create an index of all someone's messages." What exactly that covers in practice is not yet fully spelled out. TechCrunch AI reached out to OpenAI for more detail.

The plainer concern is the send button. OpenAI itself flags this one directly. The company warns users against turning on "persistent approval," a setting that would let ChatGPT send messages without asking you first each time. OpenAI's own wording: doing so "removes your final chance to review a message before ChatGPT sends it as you." That is a real risk if the chatbot misreads the tone of a conversation or replies to the wrong person.

My advice: keep the review step on. Every time.

Common questions

Is this free to use?

OpenAI has not published a separate price for the plugin specifically. ChatGPT's free tier exists, but features like this typically require a paid plan, currently ChatGPT Plus at $20 per month or a ChatGPT Work subscription. Check your account's plugin settings to see what is available at your tier.

Does ChatGPT store my messages?

OpenAI says the plugin runs locally and does not index your messages, but the full details of what data may be logged or used have not been published in full. If you handle sensitive conversations, such as medical or legal messages, it is worth waiting for clearer documentation before connecting your inbox.

Can I disconnect the plugin later?

Yes. Like any ChatGPT plugin or connected app, you can disconnect it from your account settings at any time, which stops the chatbot from accessing your messages going forward.