Dissemination level:
Public

Authors:
UL

Summary:
The Data Management Plan (DMP) of the AI4Debunk project serves as an evolving document which is periodically updated throughout the project implementation, detailing the procedure for data collection, consent procedure, storage, reuse, protection, retention and destruction of data, and confirmation that the usage and sharing of data comply with national and EU legislation. The DMP is created, confirmed, updated, and regularly monitored by the IPR working group (IPR WG) which is approved and supervised by the Innovation Management Team (IMT). The DMP contributes to the D1.1 Project Handbook, Quality Assurance Plan and Data Management Plan and D1.2 Self-assessment plan.

Keywords:
Data, plan, management, fair, use, re-use, open access, ethics.

Download here

Dissemination level:
Public

Authors:
UL

Summary:
This Self-Assessment Plan sets out the procedures in which the project’s operational performance will be assessed, including the measurement of progress toward achieving the objectives. It includes a self-assessment plan for each task within each WP from 1 to 18, recalling the objective of each task, and outlining the evaluation strategy, the success indicators, and the timetable, with the level of detail relevant at this early stage of the project. This report involves T1.1.

Keywords:
Assessment, performance, plan, tasks, objectives, progress, indicators

Download here

Dissemination level:
Public

Authors:
UL

Summary:
This document revises, complements and replaces the 1st version of the Project Handbook and Quality Assurance Plan (D1.1), additionally details assigned responsibilities, communication, meeting and reporting requirements, conflict resolution, financial requirements and incorporates the Quality Assurance Plan including instructions, procedures, checklists, and processes for reviewing deliverables and milestones. This report involves T2.1.

Keywords:
Handbook, reporting, requirements, quality assurance, procedures, processes

Download here

Dissemination level:
Public

Authors:
UL

Summary:
This is the first update of the Self-Assessment Plan, which sets out procedures for assessing project’s operational performance, including the measurement of progress toward achieving the objectives. It includes a self-assessment plan for each task within each WP from 1 to 18, recalling the objective of each task, and outlining the evaluation strategy, the success indicators, and the timetable, with the level of detail relevant at this early stage of the project. This report involves T2.2.2.

Keywords:
Assessment, performance, plan, tasks, objectives, progress, indicators

Download here

Dissemination level:
Public

Authors:
UL, CNR, P4D, EURACTIV, IUA

Summary:
The working paper establishes a theoretical framework for analyzing disinformation through a multi-dimensional lens. The paper aims to provide a comprehensive understanding of its nature, mechanisms, and far-reaching effects. Multifaceted nature of disinformation requires an analytical approach capable of capturing its sources, modes of dissemination, and the strategic intentions behind it.

Keywords:
Disinformation, disinformation threads, EU response, social media

Download here

Dissemination level:
Public

Authors:
UL, CNR, P4D, IUA

Summary:
The working paper offers a comprehensive analysis of the EU’s communication strategies, challenges, and responses amid a rapidly evolving media landscape. It examines the influence of social media, global digital platforms, and member states’ perspectives. The relevance of media literacy and critical thinking in countering disinformation is examined.

Keywords:
Critical thinking, effective instruments, EU media landscape, media literacy, social media

Download here

Dissemination level:
Public

Authors:
UL, P4D, IUA, CNR-IRPPS

Summary:
The working paper offers an in-depth examination of the societal groups most affected by foreign disinformation, the tactics and mechanisms employed by different actors, and provides actionable recommendations to strengthen resilience across sectors. Based on 43 qualitative interviews conducted in six countries and covering four key societal groups—policymakers, the business community, public opinion shapers, and the Russian-speaking diaspora—the research underscores both the breadth and depth of the challenge, while also identifying practical policy responses. Addressing this challenge requires a shift beyond traditional fact-checking toward anticipatory, collaborative, and inclusive strategies supported by innovative technological solutions like AI.

Keywords:
Disinformation tactics, sources of propaganda, target groups, threat actors.

Download here

Dissemination level:
Public

Authors:
UL, P4D, IUA, CNR-IRPPS

Summary:
The working paper reflects upon analysis of disinformation cases related to the war in Ukraine and climate change. Using a structured analytical framework, researchers examined selected case studies across various media environments to uncover recurring narratives, tactics, and actors involved in disinformation. Key findings reveal that disinformation is strategically designed to manipulate emotions, sow distrust, and reinforce societal divisions. Narratives often cluster around themes such as anti-democratic sentiments, elite conspiracy theories, and identity-based antagonism. The study also incorporated linguistic analysis of 46 Ukraine-related cases from Russian sources, revealing how language functions as a tool of deception within broader power dynamics. The findings emphasize that disinformation does not merely distort facts but reframes reality in emotionally charged and polarizing ways.

Keywords:
Climate change, critical discourse analysis, disinformation, polarizing narratives,
war in Ukraine.

Download here

Dissemination level:
Public

Authors:
MICC-UNIFI, CNIT, CNR, DOTSOFT, HU, BSC

Summary:

Report on Requirements describes and defines the requirements of the functionalities and technological solutions, implemented in WP6, WP7, WP8 and WP9, to interact with the AI4Debunk Platform. Such requirements will constitute the reference for the development of the different interfaces within WP10 and WP11.

Keywords:
Requirements, Functionalities, End-User, Interfaces, Platform

Download here

Dissemination level:
Public

Authors:
CNR, UMONS, UoG

Summary:

The deliverable D6.2 – Updated release of the dataset containing extracted features – describes the process for extracting relevant features from fake statements (e.g., topics, keywords, sentiment, and LIWC) and their related multimedia contents (e.g., captions from images, transcription from audio), including multimodal features (Meta information from body posture and gestures, and higher-level features from face recognition and voice analysis). The set of fake statements and related multimedia contents are those collected in Task 6.1 (Deliverable 6.1 Starting dataset of fake statements and related multimedia contents). The features have been extracted using the ML and multimodal AI modules developed in Tasks 8.1 and 8.2. 

Keywords:
Dataset, multimedia content, multimodal features

Download here

Dissemination level:
Public

Authors:
CNR, DOTSOFT

Summary:

The deliverable D6.3 – First report on the building process of the knowledge graphs – describes the building process of the two knowledge graphs, named the “unimodal” knowledge graph and the “multimodal” knowledge graph. The first “unimodal” knowledge graph consists in extracting the textual description from multimedia contents and adding this textual knowledge in the knowledge graph. The “multimodal” knowledge graph consists in embedding the multimedia contents within the knowledge graph based on multimodal feature description.

Keywords:
Knowledge graph, multimodal knowledge graph, Wikidata, taxonomy, ontology

Download here

Dissemination level:
Public

Authors:
UMONS, EUalive, IUA, CNR, CNIT, MICC-UNIFI, IP, HU, NUIG

Summary:

This deliverable presents a robust, scalable, and forward-looking framework for enriching the AI4Debunk knowledge graph with verified, structured, and semantically enriched data—laying the groundwork for a resilient, community-driven disinformation detection platform that can evolve and thrive beyond the initial funding period.

Keywords:
Knowledge graphs, framework, disinformation, data, platform

Download here

Dissemination level:
Public

Authors:
MICC-UNIFI, CNIT, UMONS, NUIG

Summary:

The deliverable D8.1 – Initial reports on the modules developed – describes the initial tools that have been defined and developed for debunking audio, text and image/video content. The outcome of this deliverable and of the related task T8.1 will serve as starting point for the final development of the Machine Learning based tools that will continue in T9.1.

Keywords:
Audio, text and image/video deepfake; debunking tools

Download here

Dissemination level:
Public

Authors:
UMONS, BSC

Summary:

The proliferation of multimodal news, which combines text, images, and audio, necessitates advanced systems capable of analyzing information across different formats to combat sophisticated disinformation. This report details the foundational development of the AI4DEBUNK decision support framework. We present three core components: (1) a multimodal information extraction system for audio and visual entity identification, designed to populate a central Knowledge Graph; (2) a cross-modal coherence analysis module to detect Out-of-Context (OOC) disinformation by verifying the semantic consistency between images and their captions; and (3) the initial version of the AI4DEBUNK Platform. This platform is a modular, extensible architecture that orchestrates various analysis modules—including deepfake detection, claim similarity, and coherence checking—to process multimodal news items and compute an aggregate “DisinfoScore.” While these components provide a strong foundation, future work will focus on the deep integration of the Knowledge Graph to provide the rich contextual data required to unify the system and enhance its adaptive intelligence.

Keywords:
Multimodal Disinformation, Decision Support System, Cross-Modal Analysis, Coherence Analysis, Out-of-Context (OOC), Information Extraction, Knowledge Graph, Modular Architecture, AI4DEBUNK, Fake News Detection

Download here

Dissemination level:
Public

Authors:
BSC, UMONS, CNIT, MICC-UNIFI, UoG

Summary:

As disinformation spreads rapidly, amplified by AI systems capable of generating highly convincing but false content, addressing this complex, evolving threat requires multidisciplinary and modular strategies. The AI4Debunk project responds to this challenge with a flexible, scalable architecture designed to detect and mitigate disinformation early, but making these systems trustworthy is crucial for their eventual adoption. This report details novel research on Trustworthy AI topics for disinformation detection—through causal explanations, counterfactuals, and uncertainty quantification—and proposes practical strategies for making AI4Debunk’s systems transparent, including saliency methods, heatmaps, and model cards. These efforts lay the groundwork for integrative, cross-disciplinary approaches that enhance the project’s technical robustness and societal relevance.

Keywords:
 trustworthiness, explainability, transparency, disinformation

Download here

Dissemination level:
Public

Authors:
UMONS, CNIT, MICC-UNIF, BSC, UoG

Summary:

This report details the initial implementation of the DisinfoScore (DS) aggregation mechanism, a core component of the AI4DEBUNK decision support system for combating disinformation. The platform integrates a suite of multimodal analysis modules, including deepfake detection (text, image, audio) and cross-modal coherence checking. The current methodology calculates the final DS by normalizing all activated module outputs to a [0, 1] scale and computing a uniform weighted average. This document outlines the primary limitations of this approach, namely the simplistic, nonadaptive weighting scheme and the significant challenge of system evaluation due to the lack of a suitable multimodal dataset. Future work is defined, prioritizing a migration to a more robust and flexible agent-based orchestration model to enhance modularity and introduce explainability.

Keywords:
Disinformation, Multimodal Analysis, Decision Support System, Score Aggregation, Deepfake Detection, Cross-Modal Coherence, Explainable AI (XAI)

Download here

Dissemination level:
Public

Authors:
UMONS, BSC

Summary:

The Horizon Europe-funded AI4DEBUNK project is developing an advanced decision support system to combat digital disinformation. This report details the system’s modular, multimodal platform, which integrates specialized components for comprehensive content analysis. Key modules include semantic similarity estimation, a deepfake detection suite (covering text, image, and audio), and cross-modal coherence analysis to validate image-text relationships. The platform aggregates these analyses into a DisinfoScore, a weighted average metric whose composition provides inherent explainability. We discuss this explainable-by-design architecture and outline future work, which involves transitioning to an agentic architecture orchestrated by a Large Language Model (LLM) for enhanced contextual reasoning.

Keywords:
Disinformation, Deepfake Detection, Multimodal Analysis, Explainable AI (XAI), Decision Support System, Cross-Modal Coherence, Agentic Architecture

Download here

Dissemination level:
Public

Authors:
HU, DOTSOFT, IP, UMONS, P4D, CNIT, MICC-UNIFI

Summary:

This report presents the design and development of a disinformation debunking API that can be used by software developers building user interfaces for combatting disinformation. The report outlines the API’s system architecture and how it integrates with other AI4Debunk components. It also discusses how the software will be distributed to partners and potential future users.

Keywords:
disinformation, misinformation, debunking, factchecking, web engineering, application programming interface, microservices, enterprise integration, distributed systems

Download here

Dissemination level:
Public

Authors:
HU, P4D, IP

Summary:

This report presents the design and development of a browser extension intended to help users debunk online disinformation. It reviews existing research on browser extensions that counter false information. Drawing on social science research, requirements are outlined in the form of user stories, which inform both the user interface design and system architecture. The report also details the technologies used to implement the extension and discusses how it aims to support users in identifying misleading content on the web.

Keywords:
disinformation, misinformation, debunking, browser, extension, add-on, plug-in

Download here

Dissemination level:
Public

Authors:

DOTSOFT

Summary:

AI4Debunk D10.3 specifies the AI4Debunk mobile app through detailed use cases, user stories, user journeys, and a dedicated security/privacy analysis, concluding with a technology-stack justification. The app enables multi-modal verification (text/URL, file upload, live audio, AR scanning) by sending requests to a Debunking API that returns a Disinfoscore, classification, and explanations, presented via a clear risk indicator and verdict.

Keywords:
Disinformation debunking; mobile application specification; multi-modal verification; Debunking API; Disinfoscore; onboarding and informed consent; augmented reality (AR) scanning; live audio capture; privacy and anonymity; security risk analysis; reporting to experts; Flutter cross-platform development

Download here

Dissemination level:
Public

Authors:

IP, HU, P4D

Summary:

This deliverable presents the definition, structure, and implementation design of the AI4Debunk Collaborative Platform, an interactive and participatory web environment enabling citizens, journalists, researchers, and experts to collaborate in detecting, validating, and analysing online disinformation. The platform acts as a public-facing hub within the AI4Debunk ecosystem, connecting human input with automated disinformation detection engines. It serves as an open knowledge repository (“Disinfopedia”) and as a communication interface with the browser plug-in, mobile app, and AR/VR tools developed under WP10 and WP11.

The document describes the functional and quality requirements, user experience and accessibility design, system architecture, privacy and ethical considerations, and integration with the AI4Debunk backend infrastructure. It also details the steps towards technical validation, pilot deployment, and alignment with European standards for trustworthy AI and digital inclusion.

Keywords:
disinformation, misinformation, participatory validation, collaborative platform, human-centered design, explainable AI, knowledge graph, EU Horizon Europe

Download here

Dissemination level:
Public

Authors:

DOTSOFT

Summary:

This deliverable specifies TruthSpace, AI4Debunk’s flagship VR serious game, designed to strengthen citizens’ resilience to online disinformation through immersive training. Users collaborate with the in-game AI assistant NUTRU inside a Klein-Bottle-inspired virtual world that visualises algorithmic “information loops” and overload, progressively transforming into a calm “TruthSpace” as skills are applied. The experience operationalises five core critical-thinking strategies (source evaluation, emotional manipulation recognition, fallacy/tactic detection, conspiracy framework awareness, and reflective thinking) each embodied through dedicated levels and minigames.

Keywords:
AI4Debunk; TruthSpace; Virtual Reality (VR); Augmented Reality (AR); Serious Game; Disinformation; Media Literacy; Critical Thinking; AI Assistant (NUTRU); Klein Bottle; Unity; Usability Evaluation (SUS)

Download here

Dissemination level:
Public

Authors:
Pilot4dev

Summary:
This report presents the first results of the desk research and of the online Survey of WP12. It focuses on the spread of misinformation and disinformation online, but also on the citizens’ perceptions, and also on the regulation and moderation of social media to counter the circulation of fake news.

Keywords:
Social Media Platforms, Online Survey, Citizens’ Perceptions

Download here

Dissemination level:
Public

Authors:
Pilot4dev

Summary:
This deliverable proposes the first general guidelines for the AI4DEBUNK tools’ developers. It addresses the points of User Friendliness, Social Media, Inauthentic Coordinated Behavior, Ethics, Integration of Multi-Languages, ‘explaining Fakeness’, and ‘Stakeholders integration’. The Annex II on the meeting with the beta testing group is also added.

Keywords:
Guidelines, AI tool, Inauthentic Coordinated Behaviour, Improvement, Beta testing

Download here

Dissemination level:
Public

Authors:
Pilot4dev

Summary:
This Gender Equality Plan is the first milestone towards the approval of guidelines to integrate gender equality within the different project’s work packages.

The AI4Debunk consortium is committed to including gender and intersectionality as a transversal aspect in the project’s activities. In line with EU guidelines and objectives, all partners – including the authors of this deliverable – recognise the importance of advancing gender analysis and sex-disaggregated data collection in the development of scientific research. Therefore, we commit to paying particular attention to including, monitoring, and periodically evaluating the participation of different genders in all activities developed within the project, including workshops, webinars and events but also surveys, interviews and research, in general. While applying a non-binary approach to data collection and promoting the participation of all genders in the activities, the partners will periodically reflect and inform about the limitations of their approach. Through an iterative learning process, they commit to plan and implement strategies that maximise the inclusion of more and more intersectional perspectives in their activities.

Keywords:
Women, Gender Equality, Inclusiveness, Gender Bias

Download here

Dissemination level:
Public

Authors:
Pilot4dev

Summary:
This report goes through existing literature and desk review about the case study misinformation and disinformation about climate change.

Keywords:
Climate Change, Disinformation, Misinformation, Case Study, Desk Review

Download here

Dissemination level:
Public

Authors:
Pilot4dev, UL, EUalive, IUA

Summary:
This document summarizes the findings of the stakeholders’ interviews and on the 3 focus groups organized in the year 2024, in the framework of the Task 12.5.

Keywords:
Stakeholders perceptions, Disinformation, War in Ukraine, Climate Change, Social Media Platforms

Download here

Dissemination level:
Public

Authors:
F6S

Summary:
This deliverable encompasses the initial version of the communication, dissemination and exploitation activities plan for the AI4Debunk project. It will be periodically reviewed and adjusted to meet the defined goals.

Keywords:
Communication, Dissemination, Exploitation, Strategy

Download here