Google announced the launch ofA new cloud service architecture called "Private AI Compute"This will allow users to utilize highly sophisticated AI computing functions on their devices while ensuring data privacy. The platform combines the powerful computing capabilities of high-level Gemini models in the cloud with the security and privacy guarantees that users expect from on-device processing.
This technical architecture is similar to the one Apple announced at WWDC 2024 last year.Private Cloud ComputeThe concepts are very similar, with the latter emphasizing the use of Apple Silicon processors for computing while providing more privacy-preserving "Apple Intelligence" services in the cloud and on devices.
Utilizing TPU and TIE architecture to create a "sealed" cloud environment
Private AI Compute operates on Google's own custom TPU (Tensor Processing Unit) chips and is based on a seamlessly integrated single Google stack.
Its core security mechanism lies in the integration of an architecture called "Titanium Intelligence Enclaves" (TIE), which achieves a high level of privacy and security. Through remote authentication and encryption technology, the device will connect to a hardware-protected and "sealed" cloud environment.
Google emphasizes that this means confidential data processed by Private AI Compute will be isolated from users and cannot be accessed even by Google itself, thus ensuring that the data remains private.
The Pixel 10 series is the first to adopt this feature, enabling "Magic Suggestions" and audio summaries.
The recently launched Pixel 10 series phones will be among the first new products to utilize this technology. Powered by Private AI Compute, the Pixel 10 series phones will be able to achieve more advanced AI capabilities, including:
• Magic Suggestion:Provides timely and relevant suggestions.
• Recording tool (Recorder) upgrade:Supports transcription and content summarization in a wider range of languages.
Expand bug bounties and third-party audits to increase transparency
To gain public trust, Google announced that it will include Private AI Compute in its Vulnerability Rewards Program. Furthermore, to increase system transparency, Google is allowing external audits of remote authentication and plans to expand support for the inspectability of code and binaries, accepting ongoing third-party audits.
This shows that Google has adopted the same proactive and open strategy in competing with Apple for the discourse power on AI privacy standards.
