Further Reading

An annotated reading list behind the diagram and the argument on this site. The sources are grouped by the question they help answer, rather than alphabetically, so it is easier to see how the pieces fit together. Everything cited in the About page appears here, along with additional material.

Is full-stack AI sovereignty possible?

The starting point for the Single Vertical Model on the left of the diagram: what happens when one entity tries to thread every layer of the stack at once.

• Kerry, C.F., Meltzer, J.P., & Engler, A. (2026). Is AI Sovereignty Possible? Balancing Autonomy and Interdependence. Brookings Institution & CEPS. Finds that full-stack AI sovereignty is structurally infeasible for almost any country, and proposes “managed interdependence” — strategic alliances and partnerships to reduce risk across the stack — as the realistic alternative to autarky.
• Kerry, C.F. & Mishra, S. (2025). The Myth of the Monolith: AI Is Not One Thing. Brookings. Argues that treating AI as a single technology obscures the very different policy questions raised by each layer of the stack.
• Pava, J.N., Meinhardt, C., Cryst, E., & Landay, J.A. (2026). AI Sovereignty’s Definitional Dilemma. Stanford HAI. Concludes that domestic development across the entire AI stack is prohibitively costly and unnecessary for most countries, and that the term “sovereignty” is doing too much undifferentiated work. Available at hai.stanford.edu.

Where should a small country actually build? The harness layer

If the frontier model is out of reach, the question becomes which layers are both strategically valuable and realistically achievable. The strongest recent answer points at the software around the model.

• Seger, E. & Ward-Jackson, G. (2026). Build the Roads, Not the Engine: An Open-Source Strategy for Middle-Power AI Sovereignty. Tony Blair Institute for Global Change, September 2026. Available at institute.global. The clearest statement yet of sovereignty achieved through the harness layer — the software between people and models that determines what an AI system can see, remember and do: system instructions, connections to documents and databases, tool access, agent orchestration, memory, evaluation and monitoring, and identity and permissions. The authors argue that middle powers should not race to build the world’s twentieth-best “engine,” but should build the “roads” that let increasingly abundant open models be put to work. Four recommendations follow: invest in critical open-harness infrastructure as strategic infrastructure; make interoperability and portability requirements of public-sector AI procurement (could the underlying model be swapped without rebuilding the service?); keep memory and accumulated institutional context under national control rather than inside a provider’s proprietary system; and shape the emerging rules for AI agents — especially permissions — before proprietary standards harden. The paper is a useful corrective to model-centric sovereignty debates, and maps closely onto the ecosystem model on the right of the diagram: many specialised actors building shared, interoperable infrastructure rather than one entity spanning the stack.
• Seger, E. et al. (2026). Open Source: How Middle Powers Can Build Influence in the Age of AI. Tony Blair Institute for Global Change. The companion argument that middle-power capability comes from full-stack ecosystems built on open foundations rather than frontier model ownership. Available at institute.global.
• Mozilla. (2026). The State of Open Source AI. The developer evidence underpinning the harness argument: open models are widely used but reach production less often than closed ones, because the surrounding tooling — deployment, hosting, scaling, security, compliance, maintenance — is less mature. Available at stateofopensource.ai.
• Donahoe, E. & Komaitis, K. (2026). Why AI Sovereignty Depends on Interoperability Standards. Tech Policy Press. Makes the case that standards work, not domestic ownership, is where sovereignty is won or lost.
• McBride, K., Langengen, T., West, D., Bradley, J., & Mökander, J. (2025). Sovereignty, Security, Scale: A UK Strategy for AI Infrastructure. Tony Blair Institute for Global Change. The infrastructure-side counterpart: compute, energy and data centre capacity as sovereign capability.
• The Public AI Network. (2024). Public AI: Infrastructure for the Common Good. White paper (updated October 2024). Argues for AI capability held as public infrastructure, with public accountability, rather than purely as a market product.

Mapping ecosystems, capability and agency

• World Economic Forum. (2026). Rethinking AI Sovereignty: Pathways to Competitiveness through Strategic Investments. WEF AI Global Alliance, in collaboration with Bain & Company. Source of the “ecosystem builders” archetype — smaller advanced economies that invest selectively across the AI value chain and partner internationally for the rest.
• Tech Policy Design Institute. (2025). From AI Sovereignty to AI Agency: Measuring Capability, Agency & Power. Discussion Paper. TPDi, Australia. Treats the AI ecosystem as six interdependent layers covering 101 capabilities, giving a concrete way to assess where a country actually stands.

Aotearoa New Zealand: rights, Te Tiriti and infrastructure

• Te Kāhui Tika Tangata Human Rights Commission. (2026). AI and Digital Technologies: A Human Rights and Te Tiriti o Waitangi Approach. Released 7 August 2026. Announcement at tikatangata.org.nz. The report’s central move is to treat AI and digital technologies as infrastructure rather than as discrete tools or market products — which means they require deliberate design, long-term planning and public accountability. Its first recommendation is that the Government develop a national digital infrastructure strategy anchored in human rights and Te Tiriti o Waitangi, upholding Māori data sovereignty, on a long-term planning horizon matching the country’s other infrastructure planning, and enabling digital sovereignty — with a single point of accountability to coordinate it. Its second recommendation is a national approach to human rights and Te Tiriti risk assessment and mitigation covering both public and private sector entities. Three system enablers follow: empowering people and communities (including digital inclusion, AI literacy, and support for iwi and hapū exercising self-determination over Māori data); government leadership and accountability (rights-based procurement, a public sector algorithm register, an AI harm reporting system, an AI Safety Institute); and closing legislative and regulatory gaps. It is the closest thing New Zealand has to a whole-of-society statement of what the governance layers in this diagram would need to deliver. Tom Barraclough was the principal researcher and analyst for this report.
• Barraclough, T. (2025). Sovereign AI for New Zealand. Discussion paper v3.0. Brainbox Institute. The paper underpinning this diagram: the sovereignty factors, the infeasibility of the monolith, and the case for an ecosystem built on literacy, infrastructure, fine-tuning and governance. Available at docref.org.
• Barraclough, T. (2025). Wasting Time on AI Regulation and Sovereign AI in New Zealand. Brainbox Institute. A six-part series on why New Zealand’s distributed approach to AI regulation creates information, coordination, economic and policy-direction problems. Available at brainbox.institute.
• MBIE. (2025). Artificial Intelligence Research Platform programme. New Zealand Government. The main current vehicle for domestic AI research investment.
• NZTech. (2025). Empowering Aotearoa New Zealand’s Digital Future: Our National Data Centre Infrastructure. Industry analysis of domestic compute and data centre capacity.

For outputs from this project specifically — the discussion paper, blog series, podcast and seminar materials — see the Related outputs page.

CC BY 4.0 — Tom Barraclough · tom@brainbox.institute