By 2028, more than 1.3 billion AI agents will be operating across our networks — negotiating, trading data, and executing code on our behalf.

Most of the hardware they'll run on still ships as a near-blank slate. MAC addresses, serial numbers, and soft device IDs can be cloned, spoofed, or reassigned. That is not an identity. It's a label.

At North River Labs, we built NRL Veris to fix the root of the problem: a hardware-rooted device identity derived from the physical structure of the silicon itself, using Physical Unclonable Functions. No secret is written into the chip. The secret is the chip.

For an LLM inference cluster, that means every GPU server cryptographically confirms its unique hardware-root identity before it joins the network — shrinking the attack surface in a way software-only approaches never can.