Since 2021 France's VIGINUM service has tracked foreign disinformation networks with AI tools; in 2025 it open-sourced D3lta, an LLM-based detector for coordinated inauthentic content, after exposing 1,000+ bot accounts and 43 operations around the Paris Olympics.
1,000+
Bot accounts identified in RRN network (2023-11)
43
Information-manipulation operations detected during Paris Olympics (2024)
5
Inauthentic French-language outlets exposed
~200
Pre-positioned websites in Portal Kombat network (2024-02)
Details
Maturity
Established
Promoter
VIGINUM (Secrétariat général de la défense et de la sécurité nationale)
Period
2021–2025
Keywords
national security, information integrity, digital government, open-source AI
Context
VIGINUM was created in July 2021 under France's Secretariat general de la defense et de la securite nationale (SGDSN), reporting to the Prime Minister, with a mandate to detect and characterise foreign digital interference targeting French public debate.
Activities
Built D3lta, an LLM-based tool that detects coordinated duplication of text at scale across copypasta, reworded and translated content, plus a meta-detector running ten synthetic-content detectors at once; open-sourced D3lta on GitHub in December 2024 and presented it at the Paris AI Action Summit in February 2025 so other governments, journalists and researchers could reuse and improve the method.
Results
Attributed the 'Doppelganger'/RRN influence network to the Social Design Agency and Struktura (June 2023); identified a network of over 1,000 bot accounts on X tied to RRN (November 2023); exposed five inauthentic French-language outlets; detected 43 separate information-manipulation operations during the 2024 Paris Olympics; jointly disclosed the 'Portal Kombat' network of roughly 200 pre-positioned websites with counterparts in Poland and Germany (February 2024).
Conclusions
These are published detection counts, not an independently measured reduction in citizens' exposure to disinformation; VIGINUM itself frames its tools as decision support for analysts, not automated takedown or censorship systems.
Implementation
Indicative cost
High (€500k–€5M)
Time to results
Long (> 3 years)
Staffing & skills
VIGINUM analysts (statutory government service under SGDSN/Prime Minister's office)
Conditions for success
Open-sourcing the detection tool (GitHub, December 2024) so other governments, journalists and researchers can reuse and improve the method
Statutory mandate and direct reporting line to the Prime Minister's office providing institutional backing
Combining an LLM-based coordinated-duplication detector with a meta-detector running multiple synthetic-content detectors simultaneously
Data sources
Where this practice's information was retrieved from, and when.
The UK's Government Digital Service piloted GOV.UK Chat, a retrieval-augmented generative-AI assistant answering citizens' tax, benefits and visa questions — …