How to Ensure Data Safety With Gemini AI Assistant, Google Explains

Google has started sending emails to more Android users to inform them that Gemini is replacing Google Assistant. The Mountain View-based tech giant began the official retirement of the older voice assistant in March, but in the first couple of waves, only newer devices were targeted. Now, as the company expands the Gemini AI assistant to more smartphones, it is also explaining how it manages user data from various on-device sources, and how users can take more control of their data.

What Can the Gemini AI Assistant Do?

Off the bat, the biggest advantage of Gemini as the on-device assistant is its conversational interface. Users are no longer restricted to command-based actions, and instead, they can explain what they want in natural language. Even if the prompt is vague, the chatbot can understand the intent and carry out the action.

The conversational interface allows users to access the full range of its chatbot capabilities. Additionally, the “Apps” feature can connect to various first- and third-party apps. Once connected, Gemini can not only answer queries about these apps but also perform certain tasks within them. Additionally, with screen-sharing and video-conversation features, users can get assistance with the content on their screen or around them with Gemini Live.

Gemini AI Assistant: The Big Question on Data Safety

Gemini's diverse range of capabilities also comes with a data safety concern. While the AI assistant can help out with a lot more things than Google Assistant could, it also needs to connect with more data hubs to gain context and information about the queries. Some of the apps it can connect with include Workspace apps (Gmail, Calendar, Drive, etc), Messages, Phone, WhatsApp, Spotify, YouTube, and others.

Apart from this, conversation data and the data stored during the processing of Gemini Live's screen share and video feed share features are also collected by the company.

This means a heavy Android user who relies on Gemini for everyday tasks enables the AI chatbot to handle a vast amount of data, some of which could be sensitive or private. So, how does Google handle the security of this data?

Google Reveals Its Data Handling Practices

In an email viewed by Gadgets 360 staff members, Google mentioned, “Your chats, what you share with Gemini (like files, videos, screens, and photos), audio, transcripts, and recordings of your Gemini Live interactions, your feedback, info from websites you visit with Gemini, product usage, and location info are saved in Gemini Apps Activity.

Notably, this data is reviewed by the company's trained employees and is used to improve Google services and train its AI models. These are also used to personalise the user experience. However, to protect user privacy, the company disconnects the chats from the account before sharing them with human reviewers and service providers.

In case users are uncomfortable with this data processing and involvement of humans, they can opt out of it and turn off the service by going to their profile, selecting Gemini Apps Activity, and selecting “Turn off” under the Keep activity option. After turning off the option, new chats will not be stored in the company's servers. Google will continue to retain them for 72 hours to maintain the service and process any feedback, but they will not be used for AI training.

Managing connected apps is even easier. Just by navigating to profile > Apps, users can see all the apps Gemini is connected to. Users can toggle off any option they do not want the AI assistant to process.

Google explains that Gemini accesses only the information needed to fulfil the specific request, once it has permission to do so. Additionally, it highlights that personal content from these connected apps is never seen by human reviewers, used to show ads, or used to train the underlying generative models. Further, if Gemini connects to a third-party service, its data policies and privacy practices also apply. This means Gemini cannot do anything with the data that the app was not originally intended for.