Google is testing a new 'Additional sandbox options' feature in the Gemini Desktop app that could allow the AI to access and control files, applications, and web browsing on macOS devices with reduced permission prompts. This would enable Gemini to perform actions such as reading, creating, modifying, or deleting files outside of explicitly connected folders, and interacting with native apps. Despite this expanded access, Gemini would still prompt users before executing sensitive operations. The feature is currently hidden and not publicly available, with no official confirmation from Google. Apple may impose restrictions to limit such AI access to personal data on Macs.

If enabled, Gemini could gain extensive control over a macOS device, including access to all files and the ability to operate native applications and browse the web. This expanded access could increase the risk of unauthorized data exposure or modification if the AI behaves unexpectedly or is manipulated. However, sensitive actions require explicit user confirmation, which mitigates some risk. The feature is not yet live, and no active exploitation or vulnerabilities have been reported.

As this feature is not yet publicly available and Google has not confirmed its release, no immediate action is required. Users and administrators should monitor official communications from Google and Apple for updates on this feature and any related security controls. When the feature becomes available, carefully review and configure permission settings to limit Gemini's access according to your security requirements. Apple’s anticipated restrictions may also provide additional safeguards.