Deep news Defence

Defending against deep fake news with large language and image models


September 15 2025 - September 15 2028


Summary

About

                                                                   

Title

Defending against deep fake news with large language and image models

Acronym

DeepNewsDef

NATO Project number

G8648

Duration

36 months

Total NATO co-funding budget

379 200 EUR

ICIA NATO co-funding budget

248 485 EUR

Co-financing through UEFISCDI project no. PN-IV-P8-8.2-NATO-SPS-2025-0005

Duration

  • Start date: 15 September 2025
  • Duration: 36 months

Partners

I.C.I.A.

The Research Institute for Artificial Intelligence “Mihai Drăgănescu” ICIA of the Romanian Academy

Project summary

The project will result in a disinformation defence system, implemented as a platform, integrating multiple large language models (for Romanian and Russian languages) and image detection models, augmented by human-in-the-loop secondary defence mechanisms. Given that fake news represents a significant percentage of the news in a hybrid war situation, proper defence, especially against deep fake news is paramount. We’ll employ state-of-the-art (SOTA) resources from both Romania and the Republic of Moldova, including the Representative Corpus of the Contemporary Romanian Language (CoRoLa), large language models for the Romanian and Russian languages, complemented by large image models trained on images specific to news from the two countries.

Publications

SAG: Enhancing Domain-Specific Information Retrieval with Semantic-Augmented Graphs

Carol-Luca Gasan and Vasile Pais


2025 In Proceedings of the 14th Joint Conference on Lexical and Computational Semantics (*SEM 2025), pages 362–371, Suzhou, China. Association for Computational Linguistics.
link

Other

DeepNewsDef was mentioned at the EUROLAN-2025 Summer School on NLP in the Age of Complex Networks and Agentic AI, which took place in Timișoara, Romania, 14-21 September 2025. EUROLAN-2025 focused on recent research in natural language processing, machine learning applied to language, artificial intelligence (AI) and reasoning systems, computational linguistics and implications of AI in society, education and science, in order to familiarise participants with the theory and practise underlying Generative Artificial Intelligence Models (GAIM).. The School was primarily targeted towards students at the masters, and PhD levels, while allowing also exceptional undergraduates interested in learning about the interaction between NLP, complex networks, and agentic AI.


"RELATE – A Portal for Language Technologies" - presentation by V. Păiș on 18 September 2025


RELATE – A Portal for Language Technologies

Team

I.C.I.A. Team Members

Dr. Vasile Florian PĂIȘ

Principal Investigator

Scientific Researcher II

Acad. Dan Ioan TUFIȘ


Scientific Researcher I

Dr. Verginica BARBU MITITELU


Scientific Researcher II

Dr. Radu ION


Scientific Researcher II

Dr. Tiberiu BOROȘ


Scientific Researcher II

Dr. Maria CARP


Scientific Researcher II

Eric CUREA


Scientific Researcher III

Stud. Carol-Luca GASAN


Stipend Recipient

PhD. Stud. Radu-Gabriel CHIVEREANU


Stipend Recipient


U.T.M. Team Members

Dr. Victoria Bobicev


University Lecturer (tenured),
Scientific Researcher (cumulative)

Dr. Mariana Rusu


University Lecturer (tenured),
Scientific Researcher (cumulative)

Dr. Daniela Istrati


University Lecturer (tenured)

PhD. Stud. Rodica Braniște


University Lecturer

PhD. Stud. Vadim Struna


University Lecturer

MSc. Stud. Maria Smarcevscaia