The Source Code of Law: AI Agents Rewrite and Red-Team 30,000 Articles of French Law

Turning a political programme into enforceable law means amending a vast, interconnected corpus: statutes, decrees, tax annexes and future versions that often contradict one another. We study whether a policy-configurable pipeline of AI agents can make this translation coherent, verifiable and far cheaper than manual drafting. Case study: five major French codes in full (Labour, the General Tax Code and its four annexes, Social Security, Town Planning, Construction and Housing), 30,317 articles from the official LEGI database as of 18 October 2026, rewritten according to reforms inspired by David Lisnard's 2027 programme, in an independent project. The work ran in three passes and an open public review: a first draft of every article (14,115 rewritten, 16,199 retained with a justification), a complete coherence re-reading, then a separate red team that applied 1,285 corrections and measured reliability with blind random audits. On two fresh samples totalling 1,264 articles, audited blind on both texts with the same protocol, the rewritten law has 54% fewer gross defects than the law in force (95% CI 40-64%): 3.5% ± 1.0 against 7.5% ± 1.5 (paired test, p ≈ 2×10^-10). The project cost about US$1,400 in cash, against an order-of-magnitude estimate of €6 to 36 million for the same work by lawyers. The programme is a separate input: the coherence checks do not depend on it, and we invite other programmes to be run through the same pipeline. These results, produced and measured by AI agents alone, are released for verification by lawyers and citizens. Bilingual edition: English, followed by the French version from page 22. Code and data: https://github.com/AlexMasmej/liberer-la-france Corresponding author: Alex Masmej (@AlexMasmej on X): https://x.com/AlexMasmej. Original announcement thread (19 October 2026): https://x.com/AlexMasmej/status/2112088220804722917

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-16
DOI
https://doi.org/10.5281/zenodo.24276258
Primary Topic
Artificial Intelligence in Law
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

The Source Code of Law: AI Agents Rewrite and Red-Team 30,000 Articles of French Law

Nathalie Gouailhardou, Kevin Primicerio, Alexandre Masmejean
Zenodo (CERN European Organization for Nuclear Research)
Artificial Intelligence in Law
preprint

The Source Code of Law: AI Agents Rewrite and Red-Team 30,000 Articles of French Law

Nathalie Gouailhardou, Kevin Primicerio, Alexandre Masmejean
preprint en

Abstract

Turning a political programme into enforceable law means amending a vast, interconnected corpus: statutes, decrees, tax annexes and future versions that often contradict one another. We study whether a policy-configurable pipeline of AI agents can make this translation coherent, verifiable and far cheaper than manual drafting. Case study: five major French codes in full (Labour, the General Tax Code and its four annexes, Social Security, Town Planning, Construction and Housing), 30,317 articles from the official LEGI database as of 18 October 2026, rewritten according to reforms inspired by David Lisnard's 2027 programme, in an independent project. The work ran in three passes and an open public review: a first draft of every article (14,115 rewritten, 16,199 retained with a justification), a complete coherence re-reading, then a separate red team that applied 1,285 corrections and measured reliability with blind random audits. On two fresh samples totalling 1,264 articles, audited blind on both texts with the same protocol, the rewritten law has 54% fewer gross defects than the law in force (95% CI 40-64%): 3.5% ± 1.0 against 7.5% ± 1.5 (paired test, p ≈ 2×10^-10). The project cost about US$1,400 in cash, against an order-of-magnitude estimate of €6 to 36 million for the same work by lawyers. The programme is a separate input: the coherence checks do not depend on it, and we invite other programmes to be run through the same pipeline. These results, produced and measured by AI agents alone, are released for verification by lawyers and citizens. Bilingual edition: English, followed by the French version from page 22. Code and data: https://github.com/AlexMasmej/liberer-la-france Corresponding author: Alex Masmej (@AlexMasmej on X): https://x.com/AlexMasmej. Original announcement thread (19 October 2026): https://x.com/AlexMasmej/status/2112088220804722917

Zenodo (CERN European Organization for Nuclear Research)
Peace, justice, and strong institutions
Artificial Intelligence in Law
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon