Structural Insights
What the data shows.
Analysis drawn from the Provectio index of the EU AI Act. Every number is a measurement, not an estimate.
We Measured Six GPAI Model Cards Against Annex XI. None Passed.
GPT-5, Claude Opus 4.5, Gemini 3 Pro, Grok 4, Llama 3.1, and Magistral. Six publicly available model cards from the six largest GPAI providers, measured against the 14 requirements of EU AI Act Annex XI. The highest scorer reached 89%. None achieved full coverage.
April 2026 · 7 minute read
Seven Structural Weaknesses in the EU AI Act
35 instruction gaps, 38 force deltas, 2,203 structural tensions, a fixed FLOP threshold, an undefined open-source exemption, and six pending delegated acts. The regulation's binding force is undermined by its own text in seven measurable categories.
April 2026 · 10 minute read
35 Obligations That Don't Tell You How to Comply
The EU AI Act carries 205 articles with high compulsory force. 35 of them pair that force with weak operational specification. They state that the provider must act, but not what the provider must do. These instruction gaps are where legal interpretation will diverge and the cost of guessing wrong is highest.
April 2026 · 8 minute read
23 Internal Tensions in the EU AI Act
2,203 structural tensions detected across the Act's 632 addressable units derived from 113 articles. Where one provision pulls against another, compliance teams face a choice the Act itself does not resolve.
April 2026 · 6 minute read
High-Risk Classification Under the EU AI Act
The Act defines four risk tiers. The applicable tier determines which articles bind the provider, which penalties apply, and what documentation is required before 2 August 2026.
April 2026 · 5 minute read