Building Solon: Can an AI Learn to Follow Biological Risk Through the Law?

Applied Research Update — can an AI learn to follow biological risk through the law?

Over the past several weeks, One Health Security has been looking at a deceptively simple question: what happens to governance when a biological threat refuses to remain inside the institutional category assigned to it? A pathogen may begin in an animal, infect a human, appear in wastewater, cross an international border, enter a genomic database, become the starting point for a vaccine and eventually cause governments to restrict international trade. Biologically, this can be one continuous event. Legally, it is anything but continuous.

Our Rules of Outbreaks series has followed that problem through the International Health Regulations, the new pandemic-emergency provisions, the WHO Pandemic Agreement, Pathogen Access and Benefit Sharing, the Convention on Biological Diversity and Nagoya Protocol, international trade law and the relationship between WHO and national sovereignty. That work has led to a more practical research question: could we build an AI system capable of following the biological event across those legal and institutional boundaries? We are calling it Solon.

Why Solon?

The name comes from Solon of Athens — the statesman, poet and lawgiver traditionally associated with the great legal and constitutional reforms of Athens in the early sixth century BCE. Solon inherited neither a simple problem nor an empty legal system: Athens already possessed laws and political institutions, but they were operating in a society experiencing profound economic and political strain, and his importance lies partly in attempting to reform the system rather than merely adding isolated rules to it. That makes the name appropriate for what we are trying to explore. Solon, the AI, is not intended to be an automated lawyer, and certainly not an artificial legislator; it is intended to examine relationships between rules, institutions, evidence and biological events. The distinction matters, because one of the findings emerging repeatedly from One Health Security is that biological governance rarely fails because absolutely no rule exists. More often, different rules govern different parts of the same problem. Solon is an experiment in whether AI can help us see those relationships.

What are we building?

At its simplest, Solon is an experimental One Health legal and governance intelligence system. Rather than asking a general-purpose AI model a question about outbreak law and accepting whatever it happens to generate, we want to create a controlled body of authoritative material through which the system can reason. The first research corpus is being built around The Rules of Outbreaks — the legal and institutional architecture examined throughout the series: the International Health Regulations and their 2024 amendments; the Public Health Emergency of International Concern and the new pandemic emergency; the WHO Pandemic Agreement and the Pathogen Access and Benefit-Sharing system; the Convention on Biological Diversity, the Nagoya Protocol and digital sequence information; the WTO SPS Agreement, WOAH standards, Codex Alimentarius and the International Plant Protection Convention; and the associated questions of sovereignty, surveillance, notification, trade, pathogen sharing and benefit sharing.

The articles themselves provide an interpretative layer. Underneath them, Solon needs the actual sources — treaty text, regulations, official decisions, WHO documentation, WOAH standards, WTO agreements, institutional guidance, amendments and implementation material. Where appropriate, relevant cases, scholarly analysis and historical versions of rules will eventually be added as separate evidence layers, and the distinction between primary authority and commentary will be explicit.

Training Solon does not mean teaching it our opinions

This is important. We do not want Solon simply to reproduce the conclusions of One Health Security — that would create an elaborate way of asking ourselves questions and receiving our own arguments back. The articles provide hypotheses; the underlying law provides evidence; and Solon should be able to disagree with an OHS proposition where the evidence does not support it. If we argue that a governance gap exists between two legal regimes, the useful question is not whether the AI can repeat the phrase “governance gap”. It is whether it can identify what each regime actually requires, which institutions have authority, where responsibilities overlap, where neither regime clearly governs the transition, and what evidence supports that conclusion. That makes provenance central to the project.

Every answer should have a route back to authority

A legal AI system that cannot tell us why it reached a conclusion is of limited value, so Solon needs to preserve the relationship between an assertion and its source. If it says that WHO cannot order a country to impose a vaccination mandate under the International Health Regulations, the answer should be traceable to the relevant legal and official material. If it explains when a state can provisionally introduce an SPS measure despite insufficient scientific evidence, the reasoning should connect back to Article 5.7 of the SPS Agreement. If it discusses pathogen access under PABS, it must distinguish between provisions contained in the adopted Pandemic Agreement and provisions still being negotiated in the unfinished annex.

This becomes especially important because law changes. An answer that was correct in August 2026 may not be correct after a treaty enters into force, an annex is adopted, legislation is amended or an institution issues new implementing rules. Solon therefore needs to know not only what a rule says, but when that version of the rule applied.

The first challenge: follow the pathogen

One of the first tests for Solon will be deliberately simple to describe. We give it a hypothetical emerging pathogen: it is detected in livestock; a worker becomes infected; the virus is sequenced; related cases appear overseas; the sequence is shared internationally; researchers request physical samples; a manufacturer begins vaccine development; and another country restricts imports of animals and animal products. Then we ask Solon which legal and governance regimes have become relevant.

A conventional legal search might answer each part separately. Solon should be able to construct the pathway: animal-health surveillance involving national veterinary authorities and WOAH architecture; human international spread bringing the IHR into consideration; pathogen sharing raising the emerging PABS framework; genetic resources intersecting with biodiversity and access-and-benefit-sharing governance; trade restrictions bringing the SPS Agreement and WOAH standards into the analysis; and national governments retaining sovereign decision-making responsibilities throughout. The important output is not a longer list of laws. It is a map of the relationships between them — the same map our analysis of one pathogen moving through four legal systems draws by hand.

From legal retrieval to governance reasoning

This is where the research becomes more interesting. Retrieving the correct paragraph from a treaty is useful, but it is not the problem we are ultimately trying to solve. We want to ask questions such as: where does responsibility change; where do two regimes overlap; where is information required to move from one institution to another; which transition has no clearly defined owner; which decision depends on evidence held by another sector; what legal threshold triggers escalation; and what happens if the biological signal appears before the legal threshold is satisfied? These are governance questions rather than simple legal-information questions, so Solon needs to reason about the architecture around the law.

Teaching Solon about the governance gap

One Health Security has developed several concepts through its recent applied work, and these can become testable analytical lenses within Solon. Pathogen visibility asks whether we can detect the biological threat; system visibility asks whether we can see the network through which it is moving. Detection latency measures how long it takes before a signal is observed; interpretation latency concerns the time required to understand what that signal may mean; and governance latency concerns the time between relevant evidence existing and institutions connecting it sufficiently to act. Distributed biosecurity describes situations in which successful intervention depends on actions distributed outside the institutions formally responsible for managing the threat. The objective is not to programme the AI to announce that every problem represents governance latency; it is to ask whether the concept helps explain the evidence.

A practical example

Imagine that wastewater surveillance identifies an unusual antimicrobial-resistance gene. Solon might initially identify several potentially relevant domains — environmental surveillance, human public health, AMR governance, veterinary surveillance and wastewater regulation — and then ask what evidence is available: has the same resistance mechanism appeared in clinical isolates; has it appeared in livestock; which authority receives the environmental result; is there a defined escalation threshold; does legislation require communication with another authority; and what happens if no such pathway exists? The output becomes less like a legal chatbot and more like a structured governance review.

Solon should be able to say “we don’t know”

This may be one of the most important design requirements. AI systems are extremely good at producing plausible answers, but governance research requires something different. Sometimes the correct answer is that the legislation does not specify this; or that the available evidence does not establish which authority has responsibility; or that two instruments appear to overlap but we have not identified authoritative guidance resolving the interaction; or simply that we do not know yet. Uncertainty should therefore be represented explicitly. Solon should distinguish between a legal requirement, an official interpretation, established practice, a reasonable inference, OHS analysis and an unresolved question — categories substantially more useful than an AI presenting every sentence with equal confidence.

Why start with outbreak law?

Because it is an unusually difficult test. International biological governance contains almost everything that makes legal reasoning across systems challenging: different jurisdictions, international treaties, non-binding standards, national sovereignty, scientific uncertainty, changing evidence, different institutional mandates, animal and human health, environmental surveillance, commercial interests, international trade, equity, digital information and rapidly changing rules. If Solon can reason reliably across this environment, the underlying architecture may be useful much more broadly.

This is applied research, not a finished product

Solon is at an early stage. The first objective is not to launch a public AI assistant and invite people to rely on it for legal advice; it is to determine whether the approach works. We need to test retrieval, citation accuracy, version control, conflicting authority, temporal reasoning, cross-jurisdictional reasoning, uncertainty, hallucination and source hierarchy — and, most importantly, whether mapping legal relationships actually reveals governance gaps that are useful to researchers and policymakers. Some of those experiments will fail. That is part of the project.

How we will test it

The Rules of Outbreaks series gives us a useful benchmark set, because we already know the questions we want Solon to answer: who has to notify whom when a potentially international outbreak appears; what changes legally when a PHEIC is declared; how a pandemic emergency differs; what WHO can require and what remains a sovereign national decision; when a country can restrict trade because of animal disease; how provisional action under the SPS Agreement works when evidence is incomplete; where Nagoya intersects with pathogen sharing; and what is currently agreed under PABS versus what remains under negotiation. We can compare Solon’s answers against the underlying sources and against expert review, then deliberately introduce ambiguity and conflicting rules.

We also want adversarial testing. People helping us should not merely ask Solon easy questions — we want questions designed to make it fail: a plausible but false legal proposition; an obsolete version of a rule; a question spanning WHO, WOAH and WTO responsibilities; a scenario in which the first signal appears in an animal rather than a patient; a request to distinguish an international recommendation from a binding legal obligation; a question about what the law said on a particular historical date; or a request to identify what it cannot establish. Every failure tells us something about the architecture.

How people can get involved

We want Solon to develop as an open applied-research exercise around biological governance rather than as a closed demonstration, and there are several useful ways to contribute.

• Suggest a governance problem. Researchers, veterinarians, public-health practitioners, lawyers, epidemiologists, environmental scientists, farmers and others working around biological risk encounter institutional boundaries constantly. Tell us where the system feels awkward — which information is difficult to obtain, which responsibilities overlap, where rules appear contradictory, and where a decision depends on another institution. These are potential Solon test cases.
• Submit a source. If there is legislation, official guidance, a treaty, an institutional procedure, a technical standard or an authoritative decision that should form part of the corpus, tell us. Primary sources are particularly valuable. The objective is not to make the largest legal database possible; it is to create a well-governed evidence base.
• Give us a difficult question. We want questions that cross boundaries — not simply “what does Article X say?” but “what happens when Article X encounters this animal-health rule, this trade obligation and this real-world biological event?”
• Challenge an answer. If Solon reaches the wrong conclusion, we want to know — and, more importantly, why. Provide the missing authority, identify the outdated source, or explain the institutional practice the legal text does not capture. A documented correction is research data.
• Help us identify the gaps. Eventually we want to publish a structured collection of recurring governance gaps identified through the project. Not everything needs a legislative solution: some gaps may require better data, others an escalation protocol, some institutional clarification, and others may reveal genuine legal ambiguity. Understanding which type of gap exists is the first step towards designing a sensible response.

What Solon will not be

Solon will not replace lawyers; it will not provide personal legal advice; it will not decide whether governments should impose particular health measures; it will not determine whether an organisation has complied with the law; and it will not turn uncertain scientific evidence into artificial certainty. Its purpose is narrower and, we think, more interesting: Solon is intended to help us see the governance system around a biological problem.

Why build it now?

The international biological-governance landscape is changing unusually quickly. The amended International Health Regulations are now in operation; the new pandemic-emergency mechanism has yet to face its first real test; the WHO Pandemic Agreement has been adopted but is not yet in force; PABS negotiations continue; digital sequence information is changing biodiversity governance; environmental surveillance is expanding; and genomics is connecting biological events that previously appeared unrelated. The technical ability to see biological relationships is accelerating, and the governance architecture surrounding those relationships needs to catch up. That creates an unusually good moment to experiment.

From a series of articles to a research system

The Rules of Outbreaks began by asking what law applies when infectious disease crosses borders. It ended somewhere more interesting: there is no single law governing the biological event — there is a network — and the same is true of the institutions, surveillance systems and people around it. Solon is our attempt to make that network visible. The name comes from a lawgiver who worked at a moment when the challenge was not simply writing another rule, but making institutions work across competing interests and a changing society. More than two and a half millennia later, the context could hardly be more different, yet the underlying governance question is surprisingly familiar: how do we make rules designed for separate problems work as a coherent system? That is what we want Solon to help us investigate — and we would like other people to try to break it.

Get involved

If you work in law, public health, veterinary medicine, epidemiology, environmental science, food safety, biosecurity, genomics, international trade or biological research, we are interested in the awkward questions you encounter between systems. Send us a question Solon should be able to answer; a real governance gap worth investigating; an authoritative source we should include; or a scenario designed to make the system fail. The most useful contribution may be the one that proves us wrong. Get in touch here.

Questions & Answers

Related One Health Security work

Solon grows out of the Rules of Outbreaks series and its synthesis, The Law Has a One Health Problem. See also What is One Health Security?

Applied Research update, August 2026. Solon is an early-stage research project, not a product or a source of legal advice; the legal landscape it reasons about (the IHR, the WHO Pandemic Agreement and its PABS annex, and related instruments) is itself still changing.