The ‘guardrails’ built by the EU to govern AI are inadequate in both ambition and execution, and are becoming too heavy to absorb rapid changes in technology, a new study warns.
The rules have required years of negotiation and political effort, are difficult to change but difficult to remove, and have become a “functioning trap of rigidity.”
Experts said the recent changes amounted to a “partial setback” before the EU’s AI law was fully implemented. The European Union’s AI Act 2024, which was due to come into force this year, has already been replaced by the AI Simplification Act 2026.
The study says the EU’s regulatory framework is broken and incapable of achieving the Commission’s stated goal of promoting trustworthy, human-centered and rights-respecting AI.
By contrast, the US, more by accident than by design, has “regulatory strings” that can be pulled as needed, and are more binding and enforceable than the EU’s guardrails. Rules are more specific and easier to enforce.
The study, by Alison Harcourt of the University of Exeter, Claudio M. Radaelli of the European University Institute, and Philippe Train of the University of Lausanne, says the EU’s efforts to anticipate AI risks and impose blanket rules make it difficult to develop adaptive regulations. This makes AI laws difficult to enforce and limits the ability to protect human rights and public values.
The United States will intervene legally when a particular risk clearly exists or on a sector-by-area basis. The study says this approach would be more enforceable than the EU’s AI law, and would allow more room to learn from experience as rules evolve from state to state and sector to sector.
Professor Harcourt said: “The EU’s regulatory architecture is based on an ambition to anticipate the evolution of AI risks and protect the values identified by EU institutions. It is a guardrail-oriented architecture. With this clearly normative orientation, the EU seeks to impose rules. Implementing a comprehensive structure of guardrails is a difficult challenge in AI regulation. Businesses and courts are already trying to edit and change the final outcome of AI law.”
Professor Radaelli said: “The EU has chosen a design with limited room for adaptive regulation. The US has responded only in certain areas, rather than across all AI applications, in the sense of ‘guardrail’ regulation only when there is a clear danger.”
“This approach is more specific, binding, and enforcement-oriented than the EU AI law, especially in areas such as deepfakes, election integrity, content labeling, and individual rights. By relying on state regulation rather than federal regulation, this approach is well-suited to examining whether some solutions implemented at the state level have the potential for imitation and proliferation in other states. This approach provides scope for adaptive regulation.”
The study outlines how the EU has created a framework to try to pre-empt the harms of AI before they are fully understood. The United States is particularly reactive at the state level.
/Open to the public. This material from the original organization/author may be of a contemporary nature and has been edited for clarity, style, and length. Mirage.News does not take any institutional position or position, and all views, positions, and conclusions expressed herein are those of the authors alone. Read the full text here.
