Securing Your Apple Messages: Mitigating Security and Privacy Risks

OpenAI has recently introduced a new ChatGPT feature for its Mac application, which offers significant capabilities along with potential risks. The new Apple Messages plugin enables users of the ChatGPT Work app and the code-generating tool, Codex, to “search messages, catch up on conversations, draft and send replies.”

In order to utilize this tool, users are required to grant ChatGPT full access to Messages, including all stored conversations and contact information. This plugin is accessible for all ChatGPT desktop app plans on Mac devices with Apple silicon.

Once installed and authorized, ChatGPT can analyze your text history and perform tasks such as identifying to-do items, checking meeting availability, and composing replies.

By default, permission is needed before sending a text. However, users can adjust settings on a per-chat basis to allow sending without reviewing the recipients or message contents.

Although the plugin is exclusive to ChatGPT’s Mac app, messages sent will still be visible on connected iPhones and iPads, albeit interaction from secondary devices is not supported.

This plugin is intended to handle tasks that users commonly delegate. It acts like a personal assistant and gives a glimpse into the future of AI assistants. Despite its practicality, this convenience comes with potential risks for individuals and businesses.

Concerns from business owners and privacy-conscious users are valid. The Messages app contains highly personal information on iPhones and Macs. Even though ChatGPT only accesses Messages upon request, users are essentially providing complete message history and contact details to OpenAI.

In the event of a user’s ChatGPT account being compromised, an unauthorized individual could potentially access and send harmful or malicious messages to personal contacts, colleagues, and clients under the user’s identity. Additionally, individuals and organizations are entrusting OpenAI with personal data, proprietary company information, and client data.

If a breach occurs, the repercussions could extend beyond mere data leakage to recipients receiving messages falsely attributed to the user. Rectifying such harm, especially concerning confidential business or client data, can prove challenging.

AI, despite being in its early stages, is already capable of more than just answering queries. It can take actions on behalf of users with their personal information if granted permission. Users and business owners should understand these tools and weigh the benefits against potential risks.

This is not to suggest that AI should never access personal information, as it offers tangible efficiencies. Utilizing on-device applications may present a safer option. Business owners may need to collaborate with their IT teams and enhance employee training on AI usage.