ChatGPT Can Now Send Texts For You With New Apple Messages Plug-In

ChatGPT can already write your messages.

Now, it can actually send them.

OpenAI has introduced a new Apple Messages plug-in that lets ChatGPT work directly with conversations in the Messages app on Mac. The integration can search chats, summarize conversations, prepare replies, and send messages on a user's behalf.

That sounds like a small upgrade.

It isn't.

For years, AI assistants have mostly stopped at the same place: they can tell you what to do, write something for you, or generate a response—but you still have to take the final action yourself.

This changes that.

With the Apple Messages plug-in, ChatGPT can move from generating a response to interacting with the messaging system itself.

Imagine asking ChatGPT to find your conversation with someone, summarize what you discussed, draft a response, and then send it through Messages.

The AI isn't simply helping you write anymore.

It's becoming an interface between you and another application.

There is an important limitation, though. This isn't ChatGPT suddenly taking control of Messages on your iPhone. The feature is currently designed for the ChatGPT desktop app on macOS, and it works with Apple Messages on your Mac. OpenAI says the plug-in is available on all ChatGPT plans in the desktop app and can be used through ChatGPT Work and Codex.

And by default, there's still a human checkpoint.

ChatGPT asks you to approve the message and its recipients before sending it. That matters because sending a message is very different from generating one. A badly written paragraph can be edited. A message sent to the wrong person cannot simply be unsent from reality.

The bigger story, however, is what this says about the future of AI assistants.

ChatGPT started as something you talked to.

Then it became something that could search the web, analyze files, write code, create images, and work with external services.

Now it can interact with your conversations.

That progression is important because the next generation of AI isn't necessarily going to live inside one app.

It will sit across your digital life.

Instead of opening Messages, searching for a conversation, reading through dozens of texts, deciding what needs a response, writing the response, and sending it, you could increasingly give an AI one instruction and let it handle the workflow.

That's the promise of AI agents.

And messaging is an unusually personal place to demonstrate it.

Your Messages app can contain years of conversations with friends, family, colleagues, businesses, and other people. Giving an AI access to that information creates obvious convenience—but it also creates a much bigger privacy question.

Who gets to see your conversations?

What information does the AI need to process?

Where does that information go?

And how much control should an AI have over what gets sent?

Those questions become much more important when an AI can take action instead of simply answering questions.

OpenAI's own documentation shows that the integration is designed around permissions and approvals, while also warning about risks associated with persistent approval settings.

That distinction may become one of the defining issues of the agent era.

Giving an AI permission to read something is one level of trust.

Giving it permission to change something is another.

Giving it permission to send something to another person is another level entirely.

And messaging is probably only the beginning.

If an AI can interact with Messages, there's no reason to think messaging will remain the most ambitious thing these assistants can do.

The same basic idea could eventually extend across calendars, email, documents, project-management tools, shopping services, banking interfaces, and countless other applications.

Instead of learning how to operate every piece of software individually, users could simply tell an AI what outcome they want.

That is a much bigger shift than adding another chatbot feature.

It changes the role of the AI.

The assistant stops being just a source of information and starts becoming an operator.

OpenAI's move also highlights an interesting difference between traditional software and AI agents.

Traditional apps wait for you to click buttons.

AI agents are increasingly designed to understand what you want and perform multiple steps to get there.

That could make computers dramatically easier to use.

It could also make mistakes dramatically more consequential.

A wrong answer in a chatbot is frustrating.

A wrong action performed by an agent can have real-world consequences.

That's why the approval system in the Apple Messages integration matters so much.

For now, ChatGPT still gives the user the final say before sending a message by default.

But the direction is clear.

AI assistants are moving beyond “Here's what you should say.”

They're moving toward “I'll handle it for you.”

And once AI becomes capable of operating the apps we already use every day, the competition may no longer be about which chatbot gives the best answers.

It may be about which AI can actually get the most things done.

ChatGPT just took another step in that direction.

And this time, it can press send.