Raw capture â raw/tom-barraclough-sovereign-ai-policy.md
url: https://memia.substack.com/p/tom-barraclough-sovereign-ai-policy title: "Tom Barraclough: Sovereign AI policy and digital regulatory architecture" author: Ben Reid (host), Tom Barraclough (guest) date_fetched: 2026-07-25 date_published: 2026-06-23 site: Memia (Substack)
Subtitle: AI sovereignty unpacked: what small nations need to survive the intelligence explosion
Host: Ben Reid Guest: Tom Barraclough Published: June 23, 2026 Publication: Memia (Substack newsletter/podcast)
Small nations like New Zealand cannot achieve AI sovereignty through top-down government construction of foundation models. Instead, sovereignty must be understood as a multidimensional, multi-stakeholder concept spanning compute, data, training, literacy, and individual agencyâenabled by open digital regulatory infrastructure that turns law from static PDFs into structured, versioned, referenceable datasets.
Barraclough argues law functions as an "interpretive culture" rather than a deterministic runtime. Courts apply judgment shaped by shared interpretive context, not mechanical rule-following. This has profound implications for any attempt to formalize law as computable rules.
Organizations run on layered, non-networking PDFs. Barraclough cites Jack Clark's term "paper plumbing"âa broken document architecture that fails at the moment it's most needed. This is the infrastructure problem that makes law opaque to machines.
Barraclough's startup turns law and regulation into structured, granularly referenceable datasets, enabling versioning, annotation, cross-agency interoperability, and AI accessibility. This is digital regulatory infrastructure: law as API.
Bringing more people into political discourse hasn't built shared interpretive consensusâit's exposed that the "institutional software" is past its use-by date. Barraclough cites AOC's framing: "we didn't have a rules-based order. We had a pre-rules based order."
A Cambridge journal paper in progress proposes publishing law as open, versioned, linkable datasetsâroots that "filter up through all of these different digital systems." This is the infrastructure play beneath the sovereignty conversation.
Barraclough was initially skeptical of "sovereign AI" (government builds a foundation model). He developed a multidimensional framework rejecting the notion of a single government-controlled model in favor of distributed sovereignty across compute, data, training, content moderation, literacy, and individual agency.
Even China can't nail every layer of the sovereign AI stack simultaneously. Barraclough argues getting all layers right requires an improbable perfect alignment of political, economic, legal, and technical factors.
Community-owned, values-driven AI development in Te Reo MÄori, featured as the closing epilogue to Karen Hao's Empire of AI, demonstrates what distributed, multi-stakeholder sovereignty looks like in practice.
Apple's Private Cloud Compute and the startup Tinfoil offer hardware-verifiable privacy through cryptographic attestation rather than contractual assurances.
Barraclough demonstrated a live DocRef/MCP/Claude workflow that drafted a new API standard from existing regulatory corpus in an afternoonâwork that traditionally takes nine months. He notes the tension between speed and the public good of regulatory stability.
"Any large organization that requires any kind of institution or risk management process or policies is just basically made up of PDFs."
"Getting all of them just feels like this one in a million perfect shot."
"We didn't have a rules-based order. We had a pre-rules based order" (Barraclough quoting AOC).
"There's all these different interesting layers of sovereignty that don't just require it being a government."
"Agency, to be honest, is the control plane that I'm most interested in here" (Ben Reid).