OpenAI has expanded ChatGPT’s capabilities with a new Apple Messages plugin that allows the artificial intelligence chatbot to search, analyse, draft and send messages on behalf of users, pushing the technology further into everyday digital communication.
The new integration, launched on August 20, is available through the ChatGPT desktop application on Mac and allows users to connect their Messages inbox to the chatbot. It can also be used with OpenAI’s Codex and ChatGPT Work offerings.
With the plugin, users can ask ChatGPT to search their message history, analyse conversations, suggest follow-up messages, organise information and draft responses. The chatbot can also send messages for users and delete messages on their behalf.
The development marks a further shift in the role of generative AI from responding to user prompts to taking actions across other applications.
For instance, a user could ask ChatGPT to identify people they communicated with the previous day and suggest follow-up messages based on those conversations. The system can then prepare and, with the appropriate permissions, send the messages through Apple’s Messages application.
However, the expansion has raised questions about privacy and the extent of access an AI system should have to personal communications.
According to OpenAI, the plugin does not create a complete index of a user’s messages and operates locally on the user’s device. The company said ChatGPT only reads messages when the user specifically asks it to do so.
Users must nevertheless grant the application Full Disk Access to enable the feature, a permission that gives an application broad access to files on a Mac.
OpenAI has also warned users to retain oversight of messages before they are sent. The company discourages users from enabling persistent approval because, as it explained, doing so “removes your final chance to review a message before ChatGPT sends it as you.
The privacy question is particularly significant because personal messaging often contains sensitive financial, professional and personal information. While OpenAI has outlined protections around local processing and user permissions, questions remain about what information may ultimately be retained by the AI system.
The integration is part of a broader industry shift towards AI agents capable of performing tasks rather than simply generating information. Instead of asking an AI assistant how to respond to a message, users can increasingly delegate parts of the communication process to the system.
Hence, the technology could improve business productivity by helping employees search large volumes of conversations, prepare responses and manage routine communication. For individual users, however, the ability to have AI communicate directly with other people raises questions about authenticity, accountability and whether recipients should know when a message was generated or sent by an AI system.
The development also places greater emphasis on user control as AI systems gain access to increasingly sensitive applications and personal data, while the balance between convenience and privacy is likely to become an increasingly important part of the debate surrounding AI-powered digital assistants.
