Andrew Byrley
Portrait of Andrew Byrley

Andrew Byrley

AI governance researcher · U.S. Foreign Service officer

I study how AI regulation meets reality — where policy frameworks, technical systems, and institutional capacity have to work together, and what breaks when they don't.

Recently a Winter Fellow at the Centre for the Governance of AI (GovAI) in London, where I worked on EU AI Act compliance and audit infrastructure for autonomous AI agents. Before GovAI, sixteen years as a U.S. diplomat — leading technology and economic policy at embassies in Europe and West Africa, staffing senior leadership during the Biden transition, and building coalitions on digital security and democratic resilience. I trained as a robotics engineer at Georgia Tech, and that technical foundation shapes everything I do in policy.

Currentlywriting about audit infrastructure for agentic AI under EU AI Act Article 12.
Research Writing CV (PDF) Contact
§ 01   About

Engineer by training, diplomat by career, governance researcher by conviction.

Now

Co-authoring Governing the Agent Protocol Layer, a comparative analysis of what the history of internet-protocol governance at the IETF can teach the emerging governance of AI agent protocols like MCP, and designing the AI-governance-researcher track for ECONBench in collaboration with researchers at Yale.

My research focus is the operational readiness question in AI regulation: can the institutions and organizations subject to rules like the EU AI Act actually comply with them, and what infrastructure is missing when they can't? At GovAI I worked this from the technical side — Article 12 logging requirements, Article 53 documentation standards. Now I'm looking at the institutional and comparative dimensions.

Before

On leave from a sixteen-year U.S. Foreign Service career to focus on AI governance. Most recently led economic and digital policy at Embassy Riga, where I built a strategic communications plan that positioned economic resilience as central to Latvia's national security — framing that the President and Prime Minister later echoed in their own remarks.

Earlier postings: economic diplomacy in Mauritania, staff work for the Assistant Secretary for European Affairs during the Biden transition, public affairs in Belize, and, before all of that, technical surveillance countermeasures.

Why

Sixteen years of diplomacy taught me that policy succeeds or fails based on institutional capacity — whether the people and organizations responsible for carrying it out have the tools, knowledge, and infrastructure to actually do so. A master's in robotics taught me how to read technical systems with enough depth to know when that capacity is real and when it's theater. I bring both lenses to AI governance.

§ 02   Research

Current and in‑progress work.

2026
GovAI Winter Fellowship In review

Automating GPAI Compliance Documentation

The EU AI Act requires general-purpose AI providers to produce detailed technical documentation under Article 53 — but standard model cards, even from major labs, don't contain everything regulators need, particularly around energy consumption and computational resources.

I built an MCP server that extracts metadata from HuggingFace model cards and generates Article 53 documentation automatically. Validation against Microsoft Phi-4, DeepSeek-R1, and others showed the tool works — and that the underlying documentation gaps will create industry-wide compliance challenges.

Draft paper Code on GitHub
2026
Co-authored In progress

Governing the Agent Protocol Layer

A comparative analysis asking what the history of internet-protocol governance at the IETF can offer the emerging governance of AI agent protocols like MCP. We look at where the IETF analogy holds and where it strains, the case for broader representation and open process in protocol development, and how security is best built into governance structures early rather than retrofitted once a protocol is widely deployed.

With co-authors.
2026
w/ Yale researchers In progress

ECONBench — AI governance researcher tasks

Designing the AI-governance-researcher track for ECONBench, a benchmark measuring real-world productivity gains from frontier AI across professions. The goal: tasks that probe whether models actually accelerate the work governance researchers do, not the work that's easy to grade.

2026 In progress

Logging Standards for Autonomous AI Agents

What does Article 12 actually require for agent traceability? Current logging captures what an agent did, not the reasoning chain that preceded the action. I'm exploring "externalized reasoning" as a framework for audit systems that can satisfy regulatory requirements for agents operating autonomously.

§ 03   Career

Selected postings & positions.

2026 — present
2026
2022 — 2025
2020 — 2021
2016 — 2020
2015 — 2016
2010 — 2014
§ 04   Writing

Notes on policy, systems, and institutional readiness.

Substack
Read on Substack

Shorter pieces on AI governance — how regulations interact with technical reality, what institutional readiness actually requires, and the gaps that emerge between policy intent and operational capacity.

§ 05   Contact

Get in touch.

Why

I'm interested in conversations about AI governance, regulatory design, and technology policy — with researchers, labs, governments, and civil society organizations working on these problems. I'm exploring roles in AI governance and policy advisory, particularly positions that need someone who can work across technical, institutional, and diplomatic dimensions.