OpenAI's latest ChatGPT update adds a new Mac integration with unusual reach: the ability to work directly with conversations in Apple's Messages app, including reading, searching, and sending texts on a user's behalf.

The Apple Messages plugin works with iMessage, SMS, and RCS chats, and OpenAI says it is available across all plans in the ChatGPT desktop app for macOS, with support in both ChatGPT Work and Codex. The rollout instantly became one of the most discussed AI product updates of the week — see our continuing AI tools coverage for the full picture.

What the plugin does

According to OpenAI's plugin documentation, the Apple Messages plugin can read and search Messages chats on a Mac and prepare or send messages through the Messages app. It can also analyze conversation history — OpenAI's own promotional posts suggest using it to get "insights about who you talk to, and what you talk to people about."

There are limits. The plugin does not let users interact with ChatGPT remotely through Messages, and it does not work in regular ChatGPT chats — it is scoped to the desktop app on macOS, where it can be used inside ChatGPT Work and Codex workflows. That scoping is deliberate: OpenAI frames the feature as an on-device working assistant, not a chatbot you can text from your phone.

The feature currently works only on Apple silicon Macs, not Intel machines.

By default, ChatGPT sends a message only after the user approves both the message and its recipients. Users can optionally skip that confirmation on a per-conversation basis — a convenience that shifts responsibility for accuracy onto whoever enables it.

Guardrails — and a known issue

OpenAI is unusually candid about the risks. The company's plugin guide details what it calls persistent-approval risks, explains how to revoke access, and discloses a known issue involving tasks that disable approval prompts — a scenario in which automation could potentially bypass the confirmation step users rely on.

That disclosure matters because granting an AI assistant persistent access to a messaging account is a materially different proposition from chatting with a chatbot in a browser window. Messages archives can contain passwords, financial details, medical conversations, and the texture of personal relationships — precisely the kind of context that makes an assistant more useful, and more sensitive to expose.

9to5Mac, which detailed the rollout on August 20, 2026, advised users to think carefully before granting persistent approval, noting that OpenAI's own documentation flags the failure mode.

Other features in the update

The Messages plugin is the headline addition, but the same release includes several other improvements:

  • Pinned chats sync across the ChatGPT desktop app and iOS, keeping important conversations accessible between Mac and iPhone.
  • Read-only snapshots of local Codex threads can now be shared. Snapshots do not change as the original thread evolves. Links created from personal accounts are accessible to anyone with the URL, while workspace links remain limited to workspace members. OpenAI says Codex redacts known secret patterns but warns users to review snapshots because sensitive information could remain.
  • Sites co-editing lets site owners invite active members of the same workspace to co-edit a ChatGPT Site. Editors can update content and database data, save versions, and publish changes after the owner completes the first publication, while owners retain control of access, settings, analytics, and version restoration.
  • URL changes allow some site owners to change a ChatGPT-hosted address without redeploying, with the old URL redirecting automatically.

The collection reads as a broad push to make ChatGPT a collaborative workspace product — shared threads, multi-user sites, synced state across devices — rather than a solitary chat window.

How to try it

The plugin is available on all plans in the ChatGPT desktop app for macOS, and works only on Apple silicon Macs — not Intel machines. When ChatGPT is asked to send a message on a user's behalf, it prompts for confirmation before anything is sent; skipping that confirmation is optional and per-conversation. OpenAI's plugin guide covers revocation steps for users who want to cut off access afterward.

Privacy questions and Apple tensions

The move puts OpenAI deeper onto Apple's turf at a sensitive moment. Bloomberg reported that the plugin lets ChatGPT control iMessage, potentially raising Apple privacy concerns — and Gizmodo noted bluntly that it is "something Apple probably won't like."

The release also comes amid broader friction between the two companies. Earlier in August, OpenAI publicly rebutted allegations from Apple in an ongoing dispute over trade secrets, as reported by 9to5Mac.

For now, the plugin is opt-in, gated by approvals, and documented with unusual thoroughness by OpenAI. But the direction is clear: AI assistants are steadily expanding from chat windows into the operating system's most personal surfaces — mail, files, and now messages — and each step lands with both convenience and new questions about who should hold the keys to our conversations.

---

Stay Ahead of AI

Get the latest AI news, analysis, and breakthroughs — all in one place.

Read more AI news →