How Far Should AI Go in Registry Decision-Making?

The role of artificial intelligence in registries has been a topic of discussion for several years. Despite rapid technological advancement, the core challenge remains unchanged: how to harness AI’s capabilities without compromising the legal integrity and authority that underpin every register.

This perspective builds on insights that John Murray has consistently shared with the registry community over time, including at international conferences and industry workshops. His recent reflections, captured in this LinkedIn post, provide a timely contribution to an ongoing conversation that is increasingly relevant for registry leaders and policymakers.

At a global level, this challenge is becoming more visible. Benchmarks such as the Oxford Insights Government AI Readiness Index highlight a growing divide between countries investing in AI capability and those developing the governance frameworks required to deploy it responsibly. For registries, this distinction is critical.

The Core Tension: Legal Certainty vs Probabilistic Systems

Registries are founded on legal determinism. Every entry must be authoritative, traceable, and defensible in a court of law. By contrast, AI systems operate on probabilistic models, generating outputs based on patterns, likelihoods, and historical data.

This fundamental difference creates a critical point of tension. When AI is introduced into pre-registration processes, where decisions influence legal outcomes, the risk is not the technology itself, but the potential erosion of certainty and accountability.

For senior decisionmakers, this distinction is essential. The integrity of a register depends not only on efficiency, but on the ability to stand behind every decision with absolute clarity and legal confidence.

The Key Risks of AI in Pre-Registration

As registries explore the use of AI in pre-registration contexts, it is essential to recognise and actively manage a set of interconnected risks. These are not theoretical concerns, but practical challenges that can directly impact the integrity, defensibility, and trustworthiness of the register.

1. Legal determinism vs probabilistic output
   AI systems cannot provide the level of certainty required for legal decisions. Where outcomes cannot be fully justified or defended in a court of law, they should not influence registration decisions.
2. Bias and systemic legal distortion
   AI models trained on historical data may replicate and amplify past errors, embedding jurisdictional or procedural bias at scale and introducing systemic distortion into the register.
3. Opacity and lack of explainability
   Registries require full traceability and auditability. AI models that operate as “black boxes” cannot provide the level of explanation required to support regulatory scrutiny and legal challenge.
4. Liability ambiguity
   The introduction of AI raises questions about accountability. However, legal precedent is already providing clarity. A New Zealand Court of Appeal warning on automation confirmed that automated public sector decisions remain subject to judicial review, and that the responsible authority retains full accountability. In a registry context, that responsibility ultimately sits with the registrar.
5. Adversarial risk and fraud
   AI systems can be vulnerable to manipulation. Bad actors may exploit weaknesses in pattern recognition, particularly in areas such as identity verification and document authentication.

A Foundational Principle: AI as an Enabler, Not a Decision-Maker

A consistent and pragmatic principle emerges from these considerations: AI should support registry operations, not determine outcomes.

Maintaining a human in the loop is not simply a safeguard; it is a necessity. Human oversight ensures that decisions remain grounded in legal authority, supported by judgement, and aligned with regulatory obligations.

This approach is increasingly reflected in regulation. Frameworks such as the EU AI Act classify systems that influence legal or administrative outcomes as high-risk, requiring strict standards for transparency, accountability, and human oversight. For registries, this reinforces a clear direction of travel: where legal rights are involved, AI cannot operate autonomously.

Where AI Is Delivering Value Today

While caution is required in decision-making contexts, there are several areas where AI is already delivering measurable value without compromising the integrity of the register.

• Data enrichment and pre-processing
  Extracting data from documents, structuring unstructured submissions, and identifying missing or inconsistent fields. This remains the most widely adopted use case, improving efficiency while preserving control over final decisions.
• Risk scoring (non-binding)
  Identifying potentially suspicious or fraudulent submissions to support human review. Leading implementations, such as those seen in the Danish Business Authority, demonstrate how AI can enhance oversight without replacing authority.
• Workflow optimisation
  Dynamically routing and prioritising work, predicting processing times, and improving operational efficiency across registry processes.
• Compliance support (KYC / AML)
  Screening against sanctions lists, detecting unusual patterns, and supporting compliance workflows, always with human validation at the point of decision-making.

Advancing AI in Registries with Confidence

The adoption of AI in registries is not a question of if, but how. For senior leaders, the priority is to ensure that innovation aligns with the foundational principles of the registry function: trust, authority, and accountability.

Encouragingly, governments are beginning to take steps in this direction. Initiatives such as proposals for a public registry of AI systems in Canada reflect a growing recognition that transparency must sit at the centre of AI deployment in the public sector. For registries, this is a natural extension of their role as custodians of trusted information.

The insights shared by John Murray, both in this context and through his ongoing contributions to the industry, reinforce the importance of approaching AI with clarity and discipline. Technology should enhance the register, not introduce uncertainty into its operation.

A Final Reflection

As registries continue to evolve in an increasingly digital and data-driven environment, AI will play an important role in shaping future capabilities.

However, one principle must remain constant:

Authority cannot be delegated to probability.

AI can assist, inform, and optimise, but the responsibility for decision-making must always remain with the registrar.