OpenAI's New Apple Messages Plugin Lets ChatGPT Read, Draft, and Send Your Texts
OpenAI has launched an Apple Messages plugin for ChatGPT that connects a user's Messages inbox directly to the chatbot, allowing them to sort, analyze, and edit messages from within ChatGPT's interface. TechCrunch reported the launch on August 20, 2026.
The plugin does more than read messages. Users can ask ChatGPT to delete messages, draft and send messages on their behalf, or search for information buried in their message history. OpenAI published a promotional video showing a user asking the chatbot to suggest follow-up messages to contacts based on messages received the previous day. The plugin also works with Codex (OpenAI's developer tool) and ChatGPT Work, making it usable in professional contexts alongside consumer ones. A Codex changelog entry dated August 17-21, 2026 documents the integration, listing capabilities to find chats, summarize messages, prepare replies, and send messages through Messages. OpenAI Developer Docs
OpenAI told Bloomberg that the plugin runs locally on a user's machine and "doesn't create an index of all someone's messages." That is an important architectural distinction: message content is not being bulk-uploaded to a server-side database for processing. Instead, the data stays on your device and is processed there. The company also published a help-center article explaining how user data is collected and used when ChatGPT is used through Apple's integrations, along with the controls available to users over that data. OpenAI Help Center
One design choice deserves particular attention. OpenAI discourages users from enabling persistent approval for the plugin, warning that doing so "removes your final chance to review a message before ChatGPT sends it as you." OpenAI Learn Docs. In other words, the system includes a per-message confirmation gate by default, but users can disable it. The framing in OpenAI's own documentation suggests awareness that delegating send authority to a large language model, without human review of each outgoing message, introduces a meaningful risk surface, particularly given the model's capacity to misread tone or context in a conversation it did not originate.
The broader context here is a deepening integration between OpenAI and Apple's ecosystem that has been building for over two years. Apple announced at WWDC 2024 that ChatGPT integration would come to its devices later that year. Reuters. The integration began rolling out with an iOS update on December 11, 2024. Reuters. The Messages plugin extends that relationship from system-level query access into direct manipulation of a core communication app.
That extension has not been without friction. OpenAI published a page on August 3, 2026 titled "Apple is getting this wrong," addressing an Apple lawsuit, correcting claims about its employees, and sharing messages documenting what happened. OpenAI. The dispute sits alongside the plugin launch as a reminder that the two companies' partnership coexists with unresolved legal and competitive tensions.
On the enterprise side, OpenAI's ChatGPT Enterprise & Edu release notes dated August 10, 2026 announced the retirement of individual-user sync for connected apps, with new individually authorized sync connections no longer available from that date. OpenAI Help Center. Prior to that, on August 7, 2026, ChatGPT release notes announced app experience updates making it easier to bring documents and saved files into conversations. OpenAI Help Center. These changes frame the Messages plugin as part of a broader product push to deepen ChatGPT's hooks into users' daily data flows, across both personal and professional surfaces.
For technology professionals evaluating this, the practical questions are straightforward. The plugin processes message data locally rather than transmitting a full inbox to OpenAI servers, which is a meaningfully better privacy posture than a cloud-indexing approach would be. But local processing does not eliminate the concern that a language model with send and delete authority over a user's Messages inbox is operating on one of the most sensitive data surfaces on a personal device. Text messages contain authentication codes, personal conversations, financial confirmations, and health information. Granting a model the ability to read, summarize, draft, send, and delete within that surface, even with per-message approval, concentrates a substantial amount of trust in a single integration.
In this author's view, the per-message approval gate is the critical safety mechanism, and OpenAI's own documentation recommends against disabling it. Enterprises adopting ChatGPT Work should take that recommendation as a baseline policy rather than a suggestion. The tension between convenience and control is not new in enterprise messaging integrations, but the added variable of a language model's potential to misinterpret conversational context before sending a message it drafted is a failure mode that traditional email or calendar plugins did not present.
The launch positions ChatGPT as an active participant in a user's messaging workflow rather than a passive assistant queried on demand. Whether that shift proves valuable enough to justify the trust it requires is a question users and administrators will need to answer in the coming months.
