EU AI Act a ‘rigidity trap’, University of Exeter study warns

TL;DR:

• A study by academics at Exeter, the European University Institute and Lausanne finds the EU’s AI “guardrails” fall short in ambition and execution, calling them a “rigidity trap in action”.
• The 2024 AI Act has already been replaced by the 2026 AI Simplification Act before full implementation — what the authors call “partial retreat”.
• The US approach of sector-by-sector, risk-triggered “regulatory leashes” proves more concrete, binding and enforceable, the study argues.

The EU built its AI rulebook to get ahead of harms before they were fully understood. According to new research, that ambition is precisely what broke it. A study by Professor Alison Harcourt of the University of Exeter, Professor Claudio Radaelli of the European University Institute and Dr Philipp Trein of the University of Lausanne describes the EU’s framework as a “rigidity trap”: rules that took years of political effort to build, are difficult to amend, and cannot absorb rapid changes in the technology they govern.

The evidence is already on the record. The 2024 AI Act, due to come into force this year, has been superseded by the 2026 AI Simplification Act before full implementation — a “partial retreat” in the authors’ terms, with companies and courts already contesting the Act’s final outcomes.

The counterintuitive finding concerns the United States. More by accident than design, the study argues, US “regulatory leashes” — sector-specific, state-level rules pulled when a clear danger appears — have proved more concrete, binding and enforcement-oriented than the EU’s comprehensive guardrails, particularly on deepfakes, election integrity and content labelling. State-by-state variation also creates room to learn which rules work before they spread.

Vindication or warning for the UK?

The UK’s principles-based, regulator-led approach sits between the two models the study compares, and its authors’ logic cuts both ways. Avoiding a comprehensive statute has spared the UK the EU’s rigidity trap — but the US comparison suggests enforceability comes from concrete rules somewhere, not from their absence. The UK’s bet is that existing regulators can supply that specificity sector by sector.

Looking forward

With the EU already simplifying its Act before enforcement begins, the study will strengthen the hand of those arguing the UK should hold its adaptive course — and of those pressing for binding rules in specific high-risk sectors rather than a comprehensive AI statute.