Private AI compute by Google enhances Pixel security

Google has introduced Private AI Compute, a cloud-based system powering the AI features on its Pixel 10 smartphones, Silicon Angle reports.

The Private AI Compute system hosts large AI models in Google's cloud to protect user data sent from Pixel 10 phones. Running on servers with Google's custom tensor processing unit (TPU) machine learning accelerators, the system utilizes the latest TPU, Ironwood, for enhanced performance. Google has secured the servers by disabling shell access to prevent cyberattacks that exploit this capability.

Google's use of SEV-SNP technology from AMD in intermediary servers adds an extra layer of security by encrypting memory segments, mitigating side channel attacks. By routing network traffic through IP blinding relays, Google hides users' IP addresses, making cyberattacks impractical. This approach not only enhances Pixel security but also sets the stage for more advanced AI experiences, as stated by Jay Yagnik, Google's vice president of AI innovation and research.

Source: Silicon Angle