The Role of the "Knowledge Regime" in AI Governance

After years of negotiations , EU policymakers have made the transition of the Artificial Intellgence Act (AI Act) from legislative text to a operational reality. In this long process that requires continuous expertise, research and policy evaluation, academic institutions and Think-Tanks  play a crucial role which is described as the EU’s “knowledge regime. These organizations act as intermediaries between policymakers, the technology industry, and society by translating complex technological developments into accessible policyrecommendations. They evaluate the risks and opportunities of Generative AI, study the economic and social impact of foundation models and assess how regulatin may influence Europe’s global competitiveness.

According to Pierre Bourdieu

Politcal influence is not based only on money or formal power, but also on different forms of “capital,” such as expertise, credibility, and public visibility.In the context of the EU AI Act, think tanks and academic institution provide intellectual and tehnical expertise that policymakers may not always possess, helping EU institutionsto better understand the risk and opportunities of GenerativeAI. At the same time, think tanks contribute to public debateby translating complex legal and technological issues intoaccessible information for citizens, media, and policymakers.

Institution

Centre for European PolicyStudies (CEPS)

 

Main Research Focus

General-Purpose AI value chain, macroeconomic impact, innovation policy, global governance.

Position on Regulation and Competitiveness

Supports a balancedapproach to AI regulationand argues that the AI industry is more complex than a simple value chain. It also emphasizes the importance of flexible rules for General Purpose AI (GPAI) and calls for large-scale investment in AI development across Europe.

Influence Mechanism and Key Proposals

Engages in direct lobbying through task forces and expert groups workingclosely with policymakers.

Institution

European University Institute (EUI) / Cadmus

 

Main Research Focus

Legal and political theory, “Brussels Effect", technology-neutral regulation, data flows.

Position on Regulation and Competitiveness

Skeptical about the AI Act's capacity to protectfundamental rights due to its product-safety legal basis. Warns that strict rules cause transatlantic disparities.

Influence Mechanism and Key Proposals

Produces academic legal research and participates in scholarly symposiums. It critically examines“technology-neutral” regulation, arguing that it is not just a technical design choice, but also a way of delegating regulatorypower to private companiesand standard-setting organizations.

   Institutional Voices Shaping the EU AI Act

In order to paint a clear picture, it’s important to identify and classify the main institutional voices shaping the discourse on managing generative AI risks and preserving European competitiveness.

                           The 10^25 FLOPs Threshold and the Institutionalisation of Uncertainty

The EU AI Act uses a technical benchmark of 10^25 floating-point operations (FLOPs) to identify GPAI models that are considered to pose systemic risks. Very large models such as ChatGPT-4 or Google’s Gemini fall under this category due to their high computational scale, which triggers stricterregulatory requirements such as enhanced testing, monitoring, and security obligations.

However, this approach has been widely debated.

Critics argue that the size of a model does not automatically reflect how dangerous it is. A smaller model could still cause serious harm if it is used in sensitive areas like healthcare, justice, or critical infrastructure, while a very large model may be relatively safe if it is used only for scientific or limited purposes. For this reason, many scholars believe that processing power alone is an incomplete indicator of real-world risk.

A Reactive Regulation in a Rapidly Evolving Sector

Another important criticism is that this rule is reactive ratherthan forward-looking. Since the threshold is based on existing technologies, it risks becoming outdated as AI systems evolve fast. This means that regulation may always respond to past developments instead of anticipating future risks.

Maintaining European Competitiveness: The TensionBetween Innovation and Compliance



While scholars focus on risk regulation, economic think tanks such as Bruegel and Centre for European Policy Studies analyse another issue: Europe’s growing competitiveness gap in generative AI compared to the United States and China. From a normative perspective, the EU Artificial Intelligence Act is often seen as a global pioneer, supported by the idea of the “Brussels Effect,” a concept developed in research at the European University Institute, which describes how EU regulations can become international standards due to the size and influence of the Single Market. However, some researchers also warn that the Act’s strict regulatory approachmay unintentionally slow innovation and create regulatory differences between regions, potentially limiting Europe’s ability to compete in frontier AI development.

Safety Mechanisms: Regulatory Sandboxes

In order to maintain innovation with safe use, the AI Act implemented tools that allow experimetation while maintaining regulatory oversight.

Under Article 57 of the AI Act, EU Member States must establish at least one AI regulatory sandbox. These sandboxes are controlled environments where companies and researchers can develop, test, and validate AI systems beforethey are released on the market, under the supervision of national authorities. A practical example can be found in the UK’s “AI Airlockinitiative, which supports testing of clinical AI tools in a controlled environment using real medical data before deployment in hospitals.