Technology news and discussion on TalkFuse.
Google Gives Gemini Control of Mac Apps, But Its New AI Agent Comes With Risks
Google has introduced a computer-use feature in its Gemini app for Mac, allowing the assistant to operate apps, websites and local files to complete multi-step tasks. It could save time on workflows such as transferring details from PDFs into forms, but giving an AI control across applications raises privacy and security risks.
The feature is rolling out in English to adults in the US with personal paid Google accounts, using Apple Silicon Macs running macOS Sequoia 15.0 or later. Users can watch Gemini work in a separate window, interrupt it and manage app permissions; sensitive actions require confirmation. Google warns that the agent may encounter active sessions or malicious instructions in documents and websites, and that some actions may be irreversible. macOS Accessibility and Screen & System Audio Recording permissions are also required.
In short
- Gemini can operate Mac apps, websites and files to complete tasks.
- The launch is limited to paid personal accounts in the US.
- Google warns about prompt injection, privacy and irreversible actions.
Steelman: both sides, in good faith
The strongest fair case each way — we don't pick a winner.
The case for
Giving Gemini controlled access to apps could save people time on repetitive, multi-step tasks, such as transferring information from documents into forms. The separate work window, user oversight, app permissions and confirmation for sensitive actions offer ways to keep people involved while making the assistant more useful.
The case against
An agent that can inspect files and act across apps may expose private information or be misled by malicious instructions embedded in documents or websites. Broad macOS permissions and the possibility of irreversible actions mean that mistakes or compromised sessions could have consequences beyond a single app, even with safeguards.