Tom Barraclough: Sovereign AI policy and digital regulatory architecture

A Memia podcast interview where Tom Barraclough argues that AI sovereignty for small nations like New Zealand cannot mean government-built foundation models. Instead, sovereignty is multidimensional—spanning compute, data, training, literacy, content moderation, and individual agency—and distributed across stakeholders, not concentrated in the state.

Barraclough frames law as an "interpretive culture" rather than deterministic code, which complicates any attempt to formalize it for machines. The core infrastructure problem is what Jack Clark calls "paper plumbing": organizations run on layered, non-networking PDFs that make law opaque to AI systems. His startup Syncopate, through its DocRef platform, turns regulation into structured, versioned, granularly referenceable datasets—law as API.

The conversation draws on a Cambridge journal paper in progress proposing that law be published as open, linkable datasets that filter up through digital systems. Barraclough demonstrated a live DocRef/Claude workflow that drafted a new API standard from existing regulatory corpus in an afternoon—work that traditionally takes nine months—while noting the tension between AI-accelerated lawmaking and the public good of regulatory stability.

Barraclough warns of the "one-in-a-million shot" problem: getting every layer of a sovereign AI stack right simultaneously requires improbable alignment of political, economic, legal, and technical factors. As a positive model, he points to Te Hiku Media's community-owned, values-driven AI development in Te Reo Māori. On the personal sovereignty front, Apple's Private Cloud Compute and the startup Tinfoil offer hardware-verifiable privacy through cryptographic attestation rather than contractual promises.