- OpenAI launched a desktop plugin allowing ChatGPT to read, search, draft, and send messages through the native Messages application of Apple on Mac computers.
- The integration requires explicit macOS permissions, including Full Disk Access, while running local automation tools to process message data on-device.
- The rollout comes amid rising corporate tension and ongoing trade secret litigation between Apple and OpenAI.
OpenAI released a major update for its macOS desktop application. The new feature gives ChatGPT direct control over the native Messages application of Apple.
This development lets the artificial intelligence tool search, draft, and send texts. Industry observers note that this expansion raises fresh privacy concerns for iPhone maker Apple.
Deep Integration with Apple Messaging Platforms
The new software plugin connects ChatGPT directly to private chat histories on desktop computers. The plugin allows a user to access their conversation records on a variety of messaging platforms, including iMessage, standard SMS and RCS.
Additionally, it can retrieve all the records from the old chats and find certain information users are looking for. The system can also compose fresh responses based on past conversation history.
Once the analysis of messages is complete, the artificial intelligence can produce automatic responses that are subject to approval by the senders. The users can utilize chat histories without any problems.
At present, the rollout of the software is only applicable to desktop computers. Integration is available for users who are using desktops powered by Apple Silicon and are running updated desktop software.
At the moment, the feature is only available in the specialized tiers of ChatGPT Work and Codex for desktop devices. Other web users are not able to use the messaging service on mobile devices. Mobile users will need to wait for future updates to see whether mobile support is present.
Stringent macOS Setup and Permission Barriers
To enable the integration, users have to go through a series of manual setup steps. The installation includes permission approvals that need to occur within the macOS. Users have to enable Full Disk Access in the system settings. This also requires the app to get permission to access contacts and run automation on the machine.
Such complex permission layers create barriers that prevent users from accidentally activating the software. OpenAI said that the plugin performs all of the processing on the computer of the user. The company said that the system does not keep a permanent archive of the user’s messages. Thus, all raw data is left on the computer of the host, even during analysis.
To execute these actions, the plugin leverages built-in AppleScript tools and macOS accessibility features. It avoids reverse-engineering private network protocols to maintain basic platform compatibility. Using built-in system automation capabilities, the messenger analyzes the visuals on the screen and performs actions within the operating system.
At the same time, security experts point to the fact that third-party access to communications between users poses certain risks. Cybercriminals can make use of such vulnerabilities – such as conduct prompt injection attacks via incoming messages. Such exploits might trick the chatbot into revealing sensitive personal records.
The Control for Default Approval and Security Risks
To minimize the incidents of accidents, OpenAI has also used manual verification in its operations. The interface users see for ChatGPT involves the confirmation box that checks for the authorization before any message leaves.
The process presupposes that users must confirm the sending of each message by clicking on the confirmation option. Thus, the verification ensures that messages are sent only with human intervention.
At the same time, users may want the software to automate the process of approvals for some conversations. However, OpenAI does not recommend permissiveness with this function. Ignoring manual verification means that messages are sent without regard for approval.
Researchers in the field of security warn that easy access allows for the occurrence of automated scams. In cases where an attacker sends a malicious message through text, the software may execute it. Thus, an individual review of the projects seems to be the best way to work safely.
Furthermore, it is necessary for users to safeguard the privacy of other participants in the chat. This is due to the fact that sending chat logs to an AI also impacts the person sending the incoming messages. All the friends and business partners of the user could possibly remain oblivious to the fact that their private exchanges are available to the AI system.
