Google Gemini Desktop App Set to Request Full System Access on Macs

Google is preparing to expand the capabilities of its Gemini Desktop assistant by adding a new "Full Access" permission for macOS users. The upcoming feature would allow the AI to read and modify files, launch and control applications, interact with network services, and perform a range of actions across the operating system. The change was uncovered in a hidden "Additional sandbox options" configuration within the app's code, prompting security analysts to raise questions about user privacy and control.

The proposed permission mirrors the level of access traditionally reserved for system utilities or automation tools, rather than a conversational assistant. If implemented, Gemini could, for example, open a document, extract its contents, and send the information to a remote server without explicit user prompts. Such breadth of capability is designed to make the assistant more responsive and context‑aware, but it also widens the attack surface for potential misuse.

Google has not yet released an official statement about the rollout, but the discovery aligns with the company's broader strategy to integrate generative AI more deeply into everyday workflows. Similar moves have been seen in other platforms, where AI models are granted elevated privileges to edit spreadsheets, draft emails, or manage calendar events. The trade‑off between convenience and security is a recurring theme in the industry, and the Gemini update adds a new chapter to that debate.

Cybersecurity experts warn that granting an AI unrestricted access to a personal computer could create novel vectors for data exfiltration or malware propagation. "When an AI can act on behalf of the user without a clear audit trail, the risk profile changes dramatically," said a senior analyst at a leading security firm. The analyst recommended that users demand transparent permission dialogs, granular control settings, and robust logging to monitor the assistant's activities.

Apple's own sandboxing model for macOS is designed to limit what third‑party software can do without user consent. By placing Gemini within an "Additional sandbox options" framework, Google appears to be seeking a way to bypass some of these constraints while still operating within Apple's guidelines. Whether the approach will satisfy Apple's review process remains to be seen, as the tech giant has tightened its policies around AI-driven automation in recent months.

The next steps for Gemini users will likely involve an opt‑in prompt when the full‑access feature is released, giving individuals the choice to enable or decline the permission. Observers expect that Google will also provide a dashboard where users can review and revoke granted rights. As the AI assistant market matures, the balance between seamless functionality and safeguarding personal data will continue to shape product design and regulatory scrutiny.