Google: Gemini AI for Android processes sensitive data locally

Google says it is taking a privacy-minded approach to the integration of AI features like the Gemini assistant on Android devices, implementing end-to-end protection to secure data in transit while keeping the most sensitive data locally on the device.

The data that reaches Google's cloud infrastructure for processing is protected by state-of-the-art encryption, access controls, and tight unauthorized access monitoring.

Gemini (formerly known as 'Bard') is a new AI assistant developed by Google, designed to integrate seamlessly with Android devices to help users with various work or personal tasks.

It can summarize news articles, ask questions about items the user sees on their screen, help create documents like resumes, write email responses, generate images based on user-provided descriptions, make plans and travel itineraries with Maps and Flights, translate text, and many more.

Gemini operates both on-device and in the cloud, depending on the task's complexity and privacy requirements.

For example, simpler tasks that involve sensitive information can be handled directly on the device using a lighter model called Gemini Nano, which can also work offline.

More complex tasks that require significant processing power not available through the device's processor are managed in Google's secure cloud infrastructure, and the results are sent back to the user.

As Gemini AI becomes more widely available to Android devices, Google has published a post emphasizing its commitment to providing advanced generative AI capabilities while ensuring user data privacy and security.

Key points in achieving this include:

• No third-parties: Gemini handles complex, personal tasks like workout routines or resumes within Google's secure ecosystem, without involving third-party AI providers.
• On-device AI: For sensitive tasks, Gemini Nano processes AI features directly on the device, even offline, ensuring privacy without cloud dependency.
• Secure cloud infrastructure: For demanding AI tasks, Google's secure cloud infrastructure is used, protecting data with advanced security measures and strict software controls.
• User control: Users can manage Gemini interactions by reviewing, pinning, or deleting chats and controlling app responses when the device is locked.
• Innovative privacy technologies: Google develops privacy-preserving technologies like sealed computing, which processes sensitive data in a secure cloud enclave, keeping it private even from Google.
• Enhanced transparency: Android's binary transparency allows verification of operating system code and ensures the integrity of software running in secure environments.

A more detailed white-paper on Google's end-to-end data protection about Gemini AI is to be published in the near future, so more technical details will be made available soon.