Foreign Information Manipulation & Interference

• Foreign Information Manipulation and Interference (FIMI) is defined by deliberate, coordinated, and agenda-driven efforts to deceive and disrupt target societies’ information environments.
• Empirical analyses show that FIMI campaigns deploy tactics such as narrative release, amplification, manipulation, and target degradation to polarize opinions and undermine trust.
• Detection methodologies leverage machine-learning and AI pipelines to analyze coordinated messaging, network patterns, and temporal signals for attributing and mitigating interference.

Foreign Information Manipulation and Interference (FIMI) comprises the deliberate, coordinated, and often state-sponsored use of digital, social, and traditional media channels by foreign actors to distort, disrupt, or subvert a target society’s information environment. Central to FIMI are the use of orchestrated, agenda-driven information actions intended to deceive, polarize, undermine social or institutional trust, and manipulate public opinion or behavior, typically while concealing the operation’s true provenance and intent (Erhardt et al., 2022, Pastor-Galindo et al., 17 Feb 2025, B. et al., 2020).

1. Conceptual Foundations and Formal Definitions

FIMI is formally defined by three interlocking characteristics: (1) intent to deceive; (2) orchestrated—often cross-platform—coordination; and (3) pursuit of a coherent agenda (Erhardt et al., 2022). Mathematically, for a set of information actions $A=\{a_1, a_2, ... a_n\}$, FIMI is the case where $\forall a\in A,$

$$Intent(a)=\text{deceptive} \land Coord(A)=\text{True} \land Agenda(A)=\text{True}$$

where $Coord(A)$ and $Agenda(A)$ are predicates indicating, respectively, central orchestration and the presence of a unifying strategic aim.

This structural definition distinguishes FIMI from both unintentional misinformation and uncoupled or serendipitous content virality. FIMI operations often conceal both their ultimate sponsorship and their objectives, and leverage multi-stage pipelines to establish credibility, inject and amplify narratives, and exploit sociotechnical affordances at scale (B. et al., 2020, Erhardt et al., 2022).

2. Operational Taxonomies and Strategy Archetypes

Empirical characterizations of FIMI, especially from analyses of large-scale election interference campaigns, have yielded rich taxonomies of tactics and procedures (Pastor-Galindo et al., 17 Feb 2025, Arroyo et al., 17 Dec 2025). A widely adopted framework is based on the DISARM Tactics, Techniques, and Procedures (TTPs), mapping incident metadata to structured, ATT&CK-style representations (González et al., 2 Apr 2025, Tseng et al., 21 Jan 2026).

Pastor-Galindo et al. (Pastor-Galindo et al., 17 Feb 2025) identify seven high-prevalence FIMI strategies in online operations:

• Narrative Release (97.5% of 80 mapped incidents): Launching original, anchor messages as reference points.
• Narrative Support (48.8%): Coordinated boosting via likes, shares, or synthetic comments.
• Narrative Amplification (42.5%): Driving virality using bots/influencers/SEO.
• Counter-Narrative Reaction (32.5%): Flooding counter-messaging or polarizing debate under opposing content.
• Narrative Manipulation (66.2%): Redirection to fabricated content, use of deepfakes, or strategic ad placement.
• Target Degradation (30%): Harassment, doxxing, and campaigns to silence opponents.
• Information Pollution (63.7%): Saturating communication channels with noise to obscure signal.

FIMI actors nearly always combine these strategies—over 92% of observed campaigns used at least two, and one-third used exactly four (typically release, pollution, manipulation, amplification). Conditional dependencies show that anchor-narrative release often prefigures target degradation and counter-reaction, while manipulation serves as a linchpin in complex, multi-agent deployments.

3. Detection Methodologies and Machine-Learning Pipelines

Detection of FIMI requires operationalizing the conceptual definition into measurable signals spanning credibility, coordination, and propaganda/agenda axes (Erhardt et al., 2022, Tseng et al., 21 Jan 2026). Modern pipelines employ:

• Data Ingestion & Normalization: Collection of posts, URLs, and off-platform data from diverse social and news sources, normalized into a unified schema (Erhardt et al., 2022).
• Narrative Detection & Tracking: Clustering based on entity co-occurrence (GNN, LDA, CTM/STMs) to identify evolving narrative arcs (B. et al., 2020, Park et al., 2022).
• Feature Extraction: Construction of features capturing publisher/user credibility, temporal and structural coordination (e.g., concurrent, near-duplicate posts via copy-pasta/translation/rewording in the three-Δ-space (Richard et al., 2023)), and application of known propaganda/agenda techniques (SemEval-2020 analogues).
• Classification: Use of supervised models on narrative-level feature vectors $f(n)$ including counts of coordinated posts, similarity measures, and source trust. Semi-supervised and unsupervised clustering supplement these when ground-truth is limited (Pastor-Galindo et al., 17 Feb 2025, Musulan et al., 2024).
• Attribution & Impact: Once a campaign is flagged, stylometric, network, and behavioral fingerprinting attempts to assign operations to threat actors (e.g., Russian IRA, Chinese state-linked clusters), and impact is measured using engagement, reach, and inferred belief shifts (Erhardt et al., 2022, Burghardt et al., 2024, Stoffolano et al., 15 May 2025).

Recent advances incorporate multi-agent frameworks, in which agentic AI components iteratively hypothesize, test, and verify DISARM techniques against large social media datasets, yielding interpretable, TTP-tagged evidence units with statistical confidence and effect size metrics (Tseng et al., 21 Jan 2026).

4. AI and Automation in FIMI Operations

FIMI actors have rapidly adopted AI technologies for both content production and campaign management (Arroyo et al., 17 Dec 2025, Musulan et al., 2024). Capabilities include:

• Generative Text, Image, Video (Deepfakes/“Flux-style” pipelines): Used in narrative manipulation, evidence fabrication, and impersonation.
• Synthetic Amplification: AI-generated personas, avatars, or coordinated “bot herds” provide large-scale interactive support for seeded narratives (Scala, 8 Jun 2026).
• Automated Microtargeting and Psychographic Segmentation: FIMI pipelines use collected feature vectors $x_i\in\mathbb{R}^d$ to partition populations for optimized messaging ($A_k$ segments), with real-time objective maximization via CTR and conversion rates (Fathaigh et al., 21 Sep 2025).
• Coordination via Dual-Use Platforms and Infrastructures: Use of VPNs, residential proxies, and intermediary “influence-as-a-service” companies allows adversaries to blend into platform traffic and evade attribution (Arroyo et al., 17 Dec 2025).

Consequently, technical defenses increasingly rely on detection schemes robust to paraphrasing (e.g., the three-Δ-space for copy-paste, translation, and rewording), hybrid human-AI annotation pipelines for narrative and tactic labeling, and cryptographic provenance techniques to anchor content authenticity (Richard et al., 2023, Arroyo et al., 17 Dec 2025, González et al., 2 Apr 2025).

5. Governance, Attribution, and Platform Policy Response

Major platform responses (Twitter’s State-Linked Information Operations, Meta’s Coordinated Inauthentic Behavior) employ a combination of manual and network-based detection, using signals including synchronized posting, profile misrepresentation, and shared technical infrastructure (Mugurtay et al., 2024). Logistic regression classifiers over country-level features (V-Dem Polyarchy, UNGA voting similarity, political stability, population, GDP) are empirically shown to predict platform take-down events:

$$\log\left(\frac{P(T=1)}{1-P(T=1)}\right) = \beta_0 + \beta_1 (\text{V-Dem}) + \ldots + \epsilon$$

with significantly higher take-down odds for large, authoritarian, anti-Western countries (Mugurtay et al., 2024).

Best practices emphasize cross-platform collaboration, open definition harmonization (e.g., state-linked vs. commercial operations), transparency via metadata release, and regular retraining of detection pipelines using new ground truth from takedown datasets (Mugurtay et al., 2024, Arroyo et al., 17 Dec 2025).

6. Advanced Theory: Narrative Dynamics, Moral-Emotional Framing, and Longitudinal Patterns

Contemporary FIMI research has shifted from surface-level “fake news” detection to in-depth modeling of narrative frames, BEND tactics (Engage, Explain, Excite, Dismiss, Distort, Distract, Dismay, Enhance), and longitudinal rhetorical strategies (Burghardt et al., 2024). Hierarchical ML frameworks—e.g., KcELECTRA in the Korean context (Kim et al., 22 Jun 2026)—classify text along three axes: foreign origin, moral-emotional framing, and target entity, with interpretable rationale spans supporting evidence-based moderation.

Longitudinal analysis reveals that FIMI campaigns—particularly those emphasizing “condemnation” rhetoric—spike around high-salience political events (elections, referenda). These moralizing frames achieve greater engagement, amplifying their impact on public discourse polarization (Kim et al., 22 Jun 2026).

7. Risk Mitigation, Human Factors, and Future Challenges

Human-centred frameworks such as the SOCMINT-IMS pipeline operationalize FIMI defense as a risk-calibrated, auditable sequence: from signal detection through IMS hypothesis formation, confidence/severity scoring, and proportional mitigation selection (Scala, 8 Jun 2026). By explicitly structuring detection around multidimensional coherence—semantic, temporal, infrastructural, cross-platform, cognitive—analysts can distinguish coordinated foreign interference from legitimate domestic dissent. Rigorous tabletop evaluation protocols benchmark decision quality, emphasizing mitigation proportionality and democratic safeguards.

Open challenges include advancing robust cross-platform AI detection, integrating provenance crypto-protocols, scaling digital-literacy and inoculation, and regularly adapting analytic taxonomies such as DISARM to account for emergent adversarial tactics (e.g., stealth bots, multimodal deepfakes) (Arroyo et al., 17 Dec 2025, Scala, 8 Jun 2026, Roy et al., 3 Jul 2025).

—

Key References:

• "Disambiguating Disinformation: Extending Beyond the Veracity of Online Content" (Erhardt et al., 2022)
• "Influence Operations in Social Networks" (Pastor-Galindo et al., 17 Feb 2025)
• "Deception and the Strategy of Influence" (B. et al., 2020)
• "Human-Centred Risk Mitigation for AI-Mediated Information Manipulation: A SOCMINT Framework Based on Information Manipulation Sets" (Scala, 8 Jun 2026)
• "Analysing Multidisciplinary Approaches to Fight Large-Scale Digital Influence Operations" (Arroyo et al., 17 Dec 2025)
• "Politics and Propaganda on Social Media: How Twitter and Meta Moderate State-Linked Information Operations" (Mugurtay et al., 2024)
• "Cross-National Information Attacks: A Two-Decade Analysis of Troll Behavior in Korea" (Kim et al., 22 Jun 2026)
• "Unmasking information manipulation: A quantitative approach to detecting Copy-pasta, Rewording, and Translation on Social Media" (Richard et al., 2023)
• "Microtargeted propaganda by foreign actors: An interdisciplinary exploration" (Fathaigh et al., 21 Sep 2025)
• "Online Influence Campaigns: Strategies and Vulnerabilities" (Musulan et al., 2024)
• "The Web of False Information: Rumors, Fake News, Hoaxes, Clickbait, and Various Other Shenanigans" (Zannettou et al., 2018)