Hardware AI Governance Lab

The Hardware AI Governance Lab (HAIGL) is an interdisciplinary initiative at the University of Oxford, hosted by the Oxford Martin AI Governance Initiative (AIGI) ¹. Its co-directors are Amro Awad and Robert Trager, and Ben Harack is its co-founder and research lead ¹.

The lab states that it treats hardware architecture as a foundational layer for embedding oversight, verification and control mechanisms ¹. It names three aims:

• It aims to turn high-level governance concepts into hardware design proposals, and then into early-stage feasibility prototypes ¹.
• It explores ways to verify how AI systems behave, and whether some governance constraints can be enforced directly in hardware ¹.
• It studies how agreements between states could be structured and implemented when the parties have limited trust in each other ¹.

The lab expects to release its first hardware governance design profile in late 2026 ¹. It lists two 2026 papers among its publications ¹:

• A design for fingerprinting all of an AI cluster's I/O without processors that both parties must trust. It aims to make it infeasible to covertly exfiltrate the results of undisclosed workloads through the tapped links ². Its first author lists the lab as an affiliation ². See Network taps and certifiers and Low-trust AI compute verification system overview.
• A survey of verifiable semiconductor manufacturing that develops a threat model spanning chip design through manufacturing. It argues that defence in depth is required, with assurance tiered by chip criticality, and that neutral third-party verification authorities can help overcome industry opacity ³. Two of its four authors list the lab as an affiliation ³.