ARTFEED — Contemporary Art Intelligence

Forensic Analysis of AI-Driven French Propaganda Campaign

ai-technology · 2026-08-18

A recent forensic analysis published on arXiv (2608.15746) examines the influence campaign pipeline powered by AI, unveiling PROPAGIA, a collection of 2,646 French propaganda articles from the Storm-1516/CopyCop initiative revealed by VIGINUM and INSIKT GROUP in 2025. This study contrasts PROPAGIA with SIPA, a set of human-authored French mainstream articles from the same timeframe. Researchers employed topic modeling, vagueness, and sentiment assessments, revealing that PROPAGIA significantly surpasses SIPA in vagueness, subjectivity, and negativity, while referencing fewer sources. They also identified prompt instruction leaks on 50 out of 84 PROPAGIA sites, including a detailed ten-point editorial guideline explaining some differences and notable cross-article redundancy. Additionally, rewriting-based detection aligns with INSIKT GROUP's attribution to the Llama 3 family, indicating potential involvement of other models. This research underscores the complexity of AI-driven influence tactics and the necessity of forensic techniques to trace their origins.

Key facts

  • Study published on arXiv (2608.15746) analyzing AI-driven influence campaign.
  • Introduces PROPAGIA corpus: 2,646 propagandist French articles from Storm-1516/CopyCop campaign.
  • Campaign disclosed by VIGINUM and INSIKT GROUP in 2025.
  • Comparison corpus SIPA: human-written French mainstream press from same period.
  • PROPAGIA exceeds SIPA in vagueness, subjectivity, and negativity; cites fewer sources.
  • Prompt instruction leaks found on 50 of 84 PROPAGIA websites.
  • Verbatim ten-point editorial specification identified.
  • Rewriting-based detection supports attribution to Llama 3 family, suggests other models involved.

Entities

Institutions

  • VIGINUM
  • INSIKT GROUP
  • arXiv

Locations

  • France

Sources