BPU — Processing Unit agent-readable instructions

The BPU is a hardware concept for evaluating policy circuits before an AI action reaches the host. The enforceable path is software today; FPGA and silicon phases remain roadmap work.

  • Software policy circuits: Current work compiles policy boundaries into software guard modules.
  • FPGA validation: Roadmap work would validate the execution model on hardware.
  • Silicon enforcement: Future ASIC work would require separate hardware evidence before public claims expand.

BPU — Processing Unit

Policy enforcement as a hardware roadmap.

The BPU is a hardware concept for evaluating policy circuits before an AI action reaches the host. The enforceable path is software today; FPGA and silicon phases remain roadmap work.

Architecture Overview

Separate the current product from the hardware roadmap.

Today the enforceable surface is software policy circuits. FPGA and silicon phases require their own evidence.

The Non-Maskable Block

Why the roadmap exists.

A physical gate would change where final policy decisions are enforced, but the public claim must stay phase-specific.

threat model

Software policy circuits

Current work compiles policy boundaries into software guard modules.

FPGA validation

Roadmap work would validate the execution model on hardware.

Silicon enforcement

Future ASIC work would require separate hardware evidence before public claims expand.

Roadmap

From software to silicon trajectory.

Software is available now. FPGA validates the execution model next. Silicon turns the same boundary into a physical gate.

roadmap trajectory

Phase 1: Software

Policy circuits compile to guardrail modules today (Φc = 1).

Phase 2: FPGA

Xilinx Ultrascale+ validation (12-18 months).

Phase 3: Silicon

Dedicated ASIC for AI accelerator cards (24-36 months).

Boundary: roadmap milestones are not current shipped hardware capabilities.

Phase 1 is available now.

Policy circuits are the current path. Hardware phases remain evidence-gated roadmap work.