Proof, attribution, and the work somebody else can check
AI-assisted work needs verifiable claims, clear attribution and accountable reasoning. A declaration from mathematics provides a useful starting point for acceptance criteria in research and commissioned analysis.
Keep development and test groups separate before comparing performance. No measured results are shown.
In June 2026 a group of mathematicians published the Leiden Declaration on Artificial Intelligence and Mathematics, after a workshop at Leiden University and with the endorsement of the International Mathematical Union. It is a short document and it is not a ban. It argues that the discipline’s core values — proof, attribution, and the pursuit of understanding rather than merely of results — are under pressure from tools that can produce a correct-looking answer without producing any of the three.
The declaration provides a professional-community perspective on verification and authorship. Its values can inform enterprise acceptance criteria, while the evidence and checking methods need to match the kind of work commissioned.
Require the deliverable to identify its evidence, the contribution of tools and people, and the reasoning its author can defend. These requirements help a reader assess AI-assisted work. Their relevance does not depend on claiming that all model-produced work loses those properties.
The translation
| In mathematics | In work you commission or produce |
|---|---|
| Proof — the result can be verified independently, step by step | Every claim traces to a source a reader can open, not to a summary of one |
| Attribution — who did what is recorded | Which parts were produced by a model, which by a person, and who checked them |
| Understanding — the result is not merely correct, it is explicable | Somebody in the organisation can defend the reasoning without re-running the tool |
AI-assisted drafting or analysis can be appropriate when its use and review are clear. Specify relevant disclosure and verification requirements in the statement of work. Establish what the supplier checked and who is accountable for the final claims.
The accountable author should be able to explain the argument, its evidence and its limits without asking a reader to trust the tool’s fluency. A later challenge needs an evidence record and a reasoned response, tailored to the decision the document supports.
What to ask for
Numeric claims need particular care because a plausible figure can be mistaken for an observed result. Link measured numbers to their source and denominator. Label constructed examples and verify their arithmetic. Reviewers should be able to distinguish the two.
The Institute requires numbers to be sourced or explicitly labelled as assumptions in a constructed example. This rule applies during drafting and review. It permits useful calculators and diagrams while preventing an illustration from being presented as empirical evidence.
Why this matters more in an enterprise than in a journal
Mathematics has an unusual advantage: a proof can, in principle, be checked by anyone with the time. Most enterprise work cannot. A market sizing, a risk assessment, a technical option analysis — these are read by people who do not have the source material, cannot rerun the reasoning, and are making a decision on the strength of the document’s coherence.
Fluent prose is insufficient evidence of factual quality, whether produced by a person or a model. Provide sources, reasoning and review records appropriate to the work. These let the reader assess the claims rather than rely on style as a substitute for verification.
What this does not tell you
The declaration is a statement of values by a professional community. It is not regulation, it binds nobody, and treating it as though it created an obligation would misrepresent it.
Models can support research and drafting when their contribution is reviewed and the limits remain visible. State relevant use, describe the checks and retain the source trail. The acceptance decision concerns the quality of the delivered work, not the mere presence or absence of a model.
Include provenance and verification in the acceptance criteria for external work. Request an explanation of unresolved claims before the report reaches a decision maker. A supplier’s response should be assessed for substance, with gaps resolved or clearly recorded.
Keep claims linked to their source and limit
- Article and claim
- An article contains claims with recorded status.
- Evidence link
- Each link names the source, locator and scope limit.
- Source
- A source retains its URL and access date; linking it does not establish that it supports a claim.
Representative structure from the design pack, not evidence of a measured deployment.