The Machine Reads the Law: 23 Internal Tensions in the EU AI Act
Most lawyers advising on the EU AI Act are reading it linearly. They start at Article 1 and work their way through, assuming that the text’s meaning is a stable, singular thread.
But the AI Act isn’t a story; it’s a system. And like any complex system, it contains internal tensions—points where two provisions pull in opposite directions, or where the "binding force" of a requirement is at odds with its practical "instruction force."
At Provectio, we measured the entire 113-article corpus of the Act using a traceable analysis process. By comparing obligation strength, scope, and addressee relationships across provisions, we identified 2,174 specific structural conflicts.
Here are the most significant tensions that will define the next decade of AI litigation in Europe.
1. The "Exclusively For" Trap (Article 2 vs. Article 5)
Article 2(4) exempts systems developed or used "exclusively for" military, defense, or national security purposes. However, Article 5 prohibits certain AI practices (like subliminal manipulation) without any explicit carve-out for security agencies.
The Tension:
- Article 2(4): High SITUATE force (0.95). It positions the entire Act outside the security sphere.
- Article 5: Maximized DEMAND force (0.98). It bans the practice categorically.
The Impact: If a dual-use AI system is used for a prohibited practice in a security context, which article prevails? The "Exclusively For" qualifier creates an interpretative boundary that is already showing high SCOPE_AMBIGUITY in our data.
2. The Preservation of Essence (Article 2 vs. The Charter)
Article 2(5) requires "adequate safeguards" for AI systems used by law enforcement.
The Tension:
This creates a FUNDAMENTAL_RIGHTS_TENSION. The Act attempts to balance regulatory exemptions with the EU Charter of Fundamental Rights. Our pipeline flagged this as a high-severity conflict because the "adequate safeguards" requirement has a high SITUATE score but a relatively low DO score.
The Impact: The law tells you that you must have safeguards, but it doesn't specify what they are. This creates a "Regulation Gap" where the legal burden is high, but the technical roadmap is non-existent.
3. Classification Ambiguity in Article 6
The single most important gate in the Act is the classification of "High-Risk" systems.
The Tension:
Our analysis identified a recursive CLASSIFICATION_AMBIGUITY in Article 6. The provision uses definitional language ("is to be considered") but functions as a conditional gate.
The Impact: Because Article 6 relies on Annex III (a list that can be amended by the Commission), the entire classification regime sits in a state of PF_DELEGATION_ANXIETY. Companies are being asked to comply with a framework whose boundaries are semantically fluid.
Why Linear Reading Fails
When a human reads these articles, the brain naturally smooths over these contradictions to create a coherent narrative. We assume the regulator "meant" for them to work together.
But a traceable measurement process does not smooth over those divergences. When one article carries much stronger compulsory force than a related exemption carries scope-setting force, the analysis flags a measurable gap. That gap is where legal risk and interpretation pressure can accumulate.
What This Means for Compliance
If you are building an AI system for the EU market, you cannot rely on a summary of the Act. You need to understand the structural tensions that apply to your specific use case.
Provectio provides a traceable map of these tensions. We do not just summarise the law; we show where the text creates measurable tension that legal teams may need to resolve.
Know exactly where you stand.
Book a Comprehensive Compliance Audit today. Two weeks. Verifiable forensic receipts. August-ready.
[Book Audit Here]