On-device AI security gains traction with hardware telemetry

Chipmakers and device manufacturers are increasingly focusing on on-device AI security to protect enterprise data as employees use AI tools. A significant portion of data entering AI systems now involves sensitive material, prompting a shift in how data loss prevention is handled. This evolution involves moving security classification from the cloud down to the silicon level, enabling efficient processing and ensuring data never leaves the endpoint, with further coverage provided by Silicon Angle.

The new approach relies on a security stack that extends from the neural processing unit through firmware and the boot sequence. Chipmakers and device manufacturers are integrating hardware telemetry into security consoles used by analysts. This integration allows for the detection of threats that target firmware and silicon, areas traditional security software may not reach. For instance, CrowdStrike has introduced an AI model that runs on Intel NPUs within Dell devices. This layered security strategy, extending from the client device to AI infrastructure, aims to provide full-stack AI security by monitoring endpoint behavior, AI models, data, inferencing, and the compute powering enterprise AI. This is also being applied to post-quantum readiness efforts, ensuring security from firmware signing to the boot sequence.

Source: Silicon Angle