Even though the native AI capabilities of smartphones are becoming more powerful every generation, there are still significant limitations of on-device AI processing when compared to cloud-based processing.
To overcome this limitation, Google today announced Private AI Compute, a new AI processing platform that uses large Gemini models from the cloud with the same security and privacy you generally expect from on-device processing.
In general, when you use a cloud AI service like Gemini or ChatGPT, all your data will be processed in a public cloud environment along with data from others. However, with Private AI Compute, data processing will be isolated and private to you. So, it can even process sensitive information that was previously limited to on-device processing. Google mentioned that your personal information, insights, and how you use them are protected by an additional layer of security and privacy compared to the current ones.
With Private AI Compute, Google is using remote attestation and encryption to connect your smartphone to a hardware-secured sealed cloud environment. While Gemini models can access this data in a protected environment, even Google itself won"t be able to access users" data that is being processed in this environment. Google highlighted that Private AI Compute is powered by Google"s own Tensor Processing Units (TPUs) that feature Titanium Intelligence Enclaves (TIE) for improved hardware-based security.
Using Private AI Compute, Magic Cue, an exclusive feature of the latest Pixel 10 phones, is getting even better. Google claims that this feature can now offer more timely suggestions when compared to the past. Also, the Recorder app on Pixel smartphones can now summarize transcriptions across a wider range of languages. Previously, Recorder was relying on an on-device Gemini model for transcriptions, so it had limitations on supported languages and accuracy.