The Leiden Declaration: How 150+ Mathematicians Are Drawing a Line Against Unchecked AI

The Five Threats Reshaping Mathematics

The declaration identifies five distinct but interconnected dangers, each striking at a foundational value of mathematical practice.

1. Unreliable and Unverifiable Proofs
Mathematics is built on proofs that can be independently verified and deeply understood. AI systems, however, produce arguments that look plausible but can contain nearly invisible errors—false proofs that humans struggle to catch . This problem is not limited to informal text generation; it also appears in formal proof systems when the underlying logic is obscured .

2. Collapse of Attribution and Rampant Copyright Violation
AI models train on published human work without consent and often fail to cite sources. The result is a systemic breakdown of credit, making it impossible to trace intellectual lineage or reward original thinkers. The declaration insists that authors must proactively search for antecedents and, whenever full attribution is not possible, explicitly state that limitation .

3. A Two-Tier System of Dependence and Inequality
As cutting-edge research becomes tied to expensive proprietary models and compute, mathematics faces a future where only well-funded labs can compete. This creates a structural inequality that undermines the field's traditionally open and meritocratic character .

4. Exaggerated Hype That Misleads Policymaking
Technology companies, driven by strong commercial incentives, overstate their tools’ mathematical abilities . They announce results on market timelines via press releases, not through peer-reviewed science, and use performance on math benchmarks as a marketing proxy for general intelligence—a claim the declaration flatly rejects . The authors urge governments to seek expert evaluation, not PR, when crafting science policy .

5. Loss of Research Autonomy
When corporate interests and technical feasibility dictate what gets studied, mathematics risks losing control of its own agenda. Research priorities shift toward short-term commercial returns rather than deep, curiosity-driven inquiry, threatening the discipline's long-term health .

What the Declaration Demands

Rather than just diagnosing problems, the Leiden Declaration prescribes specific, actionable norms for four key groups .

Individual Researchers must:

• Disclose all AI tools, prompts, and methods used in their work .
• Treat AI solely as a tool—never as a co-author. Humans bear full responsibility for correctness, attribution, and ethical use .
• Stay informed about emerging technologies and engage in public debate over their consequences .

Institutions, Journals, and Funders must:

• Develop binding policies on AI authorship, transparency, peer review, and intellectual property .
• Build legal and ethical support structures to help researchers negotiate with powerful technology companies .
• Weave the declaration’s core values into evaluation criteria and funding decisions, rather than rewarding mere computational speed .

Governments must:

• Stop taking corporate capability claims at face value and instead consult independent expert groups .
• Regulate the AI industry to prevent monopolistic control over research infrastructure .
• Invest in publicly funded, open alternatives to commercial AI tools so that access does not depend on institutional wealth .

Industry must:

• Abandon the practice of using research publications as training data without consent and allow real opt-out mechanisms for authors .
• Meet the same standards of transparency and attribution that academia demands of itself whenever industrial labs publish alongside academic partners .

The Bigger Fight: Proof, Power, and Public Trust

The Leiden Declaration is not just about mathematics. Its authors frame the struggle as a bellwether for science policy everywhere. They argue that the same AI systems producing unreliable proofs can also be weaponized for warfare and mass surveillance, and urge mathematicians to ethically evaluate their work and even withdraw from harmful projects .

The deeper warning is epistemological: when commercial timelines replace peer review, and when corporate hype drowns out expert caution, public understanding of what constitutes scientific truth gets distorted . Mathematics—a field that has long prided itself on clear, timeless standards—is now on the front line of that larger battle.

Nearly every recommendation in the declaration revolves around a single principle: transparency. Without knowing when and how AI was used, the scientific community cannot verify results, assign credit, or defend its own standards. With over 130 signatories at launch and institutional backing from bodies like the International Mathematical Union, the Leiden Declaration has already become more than a statement: it is a working draft of the norms that mathematicians believe the AI era demands .