The EU AI Act, which took effect in August 2024, was designed to protect citizens from risky AI systems while encouraging innovation and research. But a critical gap in how the law handles research exemptions is now creating an unintended consequence: AI researchers who publish their work at major conferences may inadvertently lose legal protection and face severe compliance obligations.

A new position paper from researchers analyzing the EU AI Act's impact on the scientific community reveals that the law's exceptions for research don't account for how modern AI research actually works. When researchers publish code, models, or demos alongside their papers on platforms like GitHub, Hugging Face, or Colab, they may be classified as "providers" of AI systems under the Act, triggering the same regulatory requirements that apply to companies like Meta, Google, or Anthropic.

Why Should AI Researchers Care About This Legal Gap?

The stakes are substantial. Non-compliance with the EU AI Act can result in fines reaching €35 million, according to the regulation's enforcement provisions. This isn't a theoretical risk. Major AI conferences, including ICML (International Conference on Machine Learning), have already updated their research ethics guidelines to require AI Act compliance, meaning researchers submitting papers must now navigate these obligations or risk rejection.

The problem stems from a fundamental mismatch between how the law defines AI systems and how researchers actually conduct and share their work. The EU AI Act defines an AI system very broadly, and when researchers combine AI models with other components like user interfaces or publish working demos, those creations qualify as regulated AI systems under the law. This means researchers must comply with extensive documentation, testing, and risk assessment requirements that were originally designed for commercial AI products.

What makes this particularly challenging is the Act's extraterritorial reach. Even researchers working outside the European Union must comply if their AI systems operate in or target the European market. This "Brussels effect" means that the regulation effectively applies to the global AI research community, not just those based in EU member states.

How Can Researchers Navigate the AI Act's Complex Requirements?

• Evaluate Your Research Status: Researchers should first assess whether they are dealing with an AI system or AI model regulated by the Act, and whether their actions classify them as "providers" of these systems under the law's definitions.
• Understand the Research Exceptions: The AI Act includes exceptions for scientific research, product-oriented research, and open-source development, but these exceptions have significant limitations that don't align with current publishing practices in the AI community.
• Plan Publication Strategy Carefully: Before publishing research artifacts alongside papers, researchers should consider whether releasing code, models, or demos might void the research exemptions and trigger full compliance obligations.
• Document Compliance Efforts: If compliance obligations apply, researchers should maintain detailed documentation of their risk assessments, testing procedures, and design choices, as the Act requires providers to document their systems throughout their lifetime.

The researchers behind this analysis propose several changes to make the AI Act more workable for the scientific community. They recommend that policymakers clarify how research exceptions apply to model and system releases, and they suggest that AI researchers take proactive steps to reduce compliance risk by understanding the Act's scope before publishing.

This situation highlights a broader challenge in AI regulation: laws designed to govern commercial products often don't fit the realities of academic research. The EU AI Act was drafted with good intentions, explicitly stating in its preamble that it should not interfere with scientific research. Yet the way the law is structured creates obstacles that researchers lack the time, budget, and expertise to navigate.

The position paper frames this as a starting point for dialogue between policymakers, legal scholars, and AI researchers. Without clarification or amendments, the current state of the law risks disrupting established research practices and potentially chilling innovation in the very field the EU aims to support. For now, researchers publishing at major conferences face a complex legal landscape that requires careful navigation to avoid unintended compliance violations.