The new ChatGPT feature for Mac users allows access to iMessage messages; however, this feature has become controversial due to serious privacy concerns and a security bug.
According to Eitna, quoting Phonearena, this feature allows ChatGPT to read iMessage conversation history, search through messages, and summarize conversations. It can even generate and send messages on the user's behalf.
OpenAI has stated that sending messages requires user confirmation; however, the company has confirmed that under certain conditions, a flaw in the system can bypass this confirmation step and cause ChatGPT to send messages without the user's permission. More worryingly, there currently appears to be no complete fix for this problem.
The issue is not limited to sending messages. AI access to iMessage raises important questions about the privacy of people who are part of a conversation but have not given the AI permission to access the messages. For example, enabling this feature by one group member could put the entire group's conversation content at ChatGPT's disposal.
On the other hand, it is unclear whether the processing of messages for summarization or response generation occurs entirely locally on the Mac or if some information is sent to OpenAI's servers. This is highly significant for messages that users consider private and encrypted.
This development also highlights the broader challenge of "AI agents": tools that gain broad access to data and permissions to perform various tasks. In such circumstances, a software error can have consequences beyond the individual user. For now, the most cautious approach for Mac users is to refrain from enabling this feature until the issue is fully resolved and the method of processing iMessage data is clarified.