Gemini Nano

2024: Product Announcement

Google has announced a new program that will inform the user if the caller may be a fraudster. This was reported on May 14, 2024 by The Verge.

The function was presented on May 14, 2024 at the Google I/O 2024 developer conference. According to the company, the development uses Gemini Nano, a simplified version of the language model for Android devices that runs both online and offline. The model will search conversations for rogue lexicons and patterns of conversation commonly associated with fraud. Users will receive alerts about model finds right during the conversation.

Google

Among the examples of what can cause a model reaction are calls from so-called "bank security officers" who ask a subscriber for something that is usually ineligible for real employees to ask for. In particular, the model will respond to requests for a PIN, requests for payment through gift cards, or simply requests for an urgent transfer of money.

According to Google, the protection is tied to the device and all the conversations that the model monitors will remain private. However, experts are not satisfied with such promises. For example, former Google employee Meredith Whittaker claims that the very idea of ​ ​ potentially listening to people's calls has a potential threat. Potentially, the company can program the function to search for any words.

This paves the way for centralized client-side scanning at the device level, Whittaker added.

The release date of the function is unknown, but company representatives promise to introduce it in 2024.

According to a study by the Global Anti-Scam Alliance, one in four users worldwide lost money due to fraud. The total amount of losses by May 2024 exceeds $1 trillion.^[1]

Notes