The Architecture of an All-Out AI-Driven Cognitive Warfare Campaign: Mechanisms, Vectors, and Epistemic Impacts#

The nature of strategic geopolitical conflict has irreversibly shifted. While traditional warfare centers on the control of geographic domains—land, sea, air, space, and cyber—the contemporary security environment has witnessed the emergence of a decisive sixth operational domain: human cognition. In this environment, the human brain is no longer merely the director of conflict; it is simultaneously the target and the weapon in the fight for cognitive superiority1. An all-out, outright psychological and psychological operations (PSYOPS) attack driven by artificial intelligence (AI) represents a paradigm shift from historical statecraft and military deception. It does not merely seek to influence what target populations think; it seeks to fundamentally alter the conditions under which human judgment, deliberation, and collective action become possible3. Historically, psychological warfare involved the planned use of propaganda and information support operations to influence the opinions, emotions, and behaviors of opposition groups4. In ancient times, strategists like Sun Tzu relied on deception and the manipulation of adversary leadership; by the 20th century, mass broadcast technologies expanded the target to entire populations5. However, the advent of generative AI, large language models (LLMs), and autonomous multi-agent systems has transitioned the operational framework from a broadcast model of mass persuasion to a hyper-personalized, continuously adaptive model of epistemic sabotage. As articulated in ancient texts such as the Dhammapada, human conflicts ultimately revolve around the mind; applied to modern warfare, the core message remains that the intrapersonal and interpersonal cognitive space is the ultimate battleground2. An all-out AI-driven PSYOPS campaign operates below the thresholds of conventional armed conflict, seamlessly blending strategic competition with covert subversion5. It is characterized by a continuous battlespace operating non-kinetically, expanding the target set from discrete military platforms to fundamental human cognitive and social systems, including trust networks, identity narratives, and institutional legitimacy7. This report provides an exhaustive analysis of the theoretical and operational architecture of a full-scale AI cognitive warfare campaign. By synthesizing contemporary research on multi-agent swarms, psychographic micro-targeting, cognitive immunology, and the sociological mechanics of synthetic consensus, the analysis delineates the exact mechanisms through which an adversary could achieve total cognitive dominance and epistemic paralysis over a target society.

The Tripartite Framework of Cognitive Disruption#

To understand the mechanics of an all-out AI PSYOPS attack, it is essential to disaggregate the target space. Cognitive warfare is an interdisciplinary construct, positioned at the nexus of neuroscience, behavioral psychology, and digital technology5. Advanced frameworks developed by international defense research organizations, such as the North Atlantic Treaty Organization (NATO) Allied Command Transformation (ACT) and the NATO Science & Technology Organization (STO), conceptualize cognitive engagements across a multidimensional spectrum5. The 2025 NATO Chief Scientist's Report on Cognitive Warfare explicitly frames contemporary conflict as increasingly behavior-centric, where the decisive terrain is how individuals and groups perceive, interpret, decide, and act7. In a full-scale AI attack, human cognition is targeted simultaneously across three deeply interconnected levels—Biological, Psychological, and Social—utilizing a combination of bottom-up, middle-out, and top-down operational vectors7.

The Biological Level: Manipulating Capacity#

At the foundational level, cognitive warfare directly targets the human nervous system, treating it as the primary substrate of thought, emotion, and behavior7. While physical neurotechnologies (neuroS/T) represent an emerging vector in direct combat psychology—such as the "Intelligent Psychological Monitoring System" smart bracelets developed by China to record facial information and emotional changes in soldiers—an AI-driven digital attack primarily influences the biological level through behavioral conditioning and neurochemical hijacking1. AI recommendation algorithms and automated content delivery systems on social media platforms are explicitly optimized for engagement. By their commercial design, these systems inherently privilege content that triggers high-arousal emotional states such as fear, anger, and moral outrage3. In an all-out attack, an adversary leverages this existing infrastructure. AI systems generate high-velocity, emotionally activating stimuli designed to bypass rational cognitive inhibition3. By flooding the information environment with highly stimulating, hyper-personalized content, the attack induces continuous physiological arousal and attentional saturation. This effectively degrades the target's baseline cognitive capacity, altering attentional gating mechanisms and rendering the population biologically susceptible to subsequent psychological manipulation7. The biological targeting acts as a bottom-up approach, ensuring that the target's physiological state is already compromised before they even attempt to process the semantic meaning of the propaganda.

The Psychological Level: Manipulating Interpretation#

With the target's biological capacity degraded, the attack shifts to the psychological manipulation of interpretation. The focus here is on influencing cognitive appraisals, framing, and the specific patterns of thought that govern individual beliefs and judgments7. At this level, generative AI acts as an unprecedented accelerant by generating highly persuasive, logically coherent, and syntactically flawless narratives at scale10. AI-enabled influence systems analyze vast troves of digital footprints, social graphs, and behavioral data to identify specific psychological vulnerabilities. The attack does not rely on a monolithic, broadcast propaganda narrative; instead, LLMs dynamically tailor distinct stimuli to exploit the unique volatilities of specific individuals or micro-communities7. For example, research highlights the growing importance of "gamified narrative language" and gamer slang as cognitive warfare tactics, allowing adversaries to subtly infiltrate youth demographics and subcultures using familiar, deprecating narratives rather than overt political messaging4. By leveraging fear conditioning, confirmation bias, and identity-protective cognition, the AI system shapes the target's psychological framing of reality before they possess the cognitive bandwidth to critically evaluate the information3.

The Social Level: Manipulating Cohesion#

The ultimate objective of a systemic PSYOPS campaign is realized at the social level, which serves as the overarching domain for influencing shared narratives, institutional legitimacy, and civic values7. Here, the attack scales from the individual to the population. The strategic goal is not necessarily to persuade a majority of the population to adopt a specific adversarial viewpoint—which is often too high a hurdle—but rather to fracture societal cohesion, weaponize domestic identities, and engineer what researchers term "epistemic chaos"7. An all-out attack paralyzes a society's "cognitive flexibility," defined as the capacity of a democratic nation to tolerate dissenting viewpoints, deliberate across ideological divides, and discover common ground3. Through the hyper-polarization of public discourse, democratic institutions become gridlocked, and institutional cynicism deepens to a point of critical failure3. The targeted society is thus rendered incapable of collective decision-making or coordinated crisis response. This achieves the adversary's strategic objectives—such as preventing intervention in a foreign conflict or degrading economic productivity—without the deployment of kinetic military force or the crossing of traditional thresholds of armed conflict. The measure of effectiveness shifts from short-term message penetration to durable, permanent changes in societal behavioral dispositions7.

The Dimensions of Harm and the Dangers of Over-Militarization#

In assessing these levels, it is vital to contextualize the dimensions of harm inflicted by cognitive warfare. Theorists such as Miller differentiate cognitive warfare from traditional cyber conflict by outlining four specific dimensions of harm: physical or psychological harm to humans, damage to physical objects, damage to software and data, and critically, harm to institutions5. Covert cognitive warfare subtly undermines targets without escalating to open hostilities, inflicting psychological harm on individuals while systematically damaging the structural integrity of civic institutions5. However, the classification of the cognitive space as an outright "warfighting domain" is not without significant academic and policy contention. Analysts from the Center for Security Studies (CSS) at ETH Zurich have issued stark critiques regarding the militarization of the cognitive domain. They argue that treating cognition as a warfighting domain inflates the threat and over-militarizes domestic, structural societal problems3. The diagnosis of cognitive warfare often conflates malicious input (e.g., coordinated bots and synthetic media) with strategic impact, making it difficult to separate the effects of foreign campaigns from pre-existing domestic anxieties3. By treating systemic vulnerabilities—such as polarized politics, fragmented media economies, and commercial attention capture—as direct evidence of foreign military strategy, democratic nations risk directing the problem to the wrong public bodies. Defence ministries are ill-equipped to fix issues meant for civilian institutions like competition authorities and data protection agencies3. Importing the conceptual architectures of autocratic regimes, where domestic populations are treated as operational targets and military intelligence polices the information ecosystem, severely risks securitizing legitimate political contestation and eroding the core democratic advantage of a free press3. Therefore, an all-out AI PSYOPS attack actively exploits this tension, forcing a democracy to choose between epistemic vulnerability and the authoritarian restriction of free expression.

The Vanguard of the Attack: Malicious AI Swarms and Agentic Systems#

The execution of a population-scale psychological attack relies on the deployment of agentic AI systems. Traditional disinformation campaigns, such as the use of "copy-paste" botnets or highly regimented troll farms during the late 2010s, were fundamentally limited by human labor requirements, rigid scheduling scripts, and easily detectable signatures of coordinated inauthentic behavior13. The modern era introduces a vastly more sophisticated threat: "malicious AI swarms," which fuse the advanced reasoning and generative capabilities of LLMs with autonomous multi-agent architectures12.

The Transition to Agentic Warfare#

Agentic AI represents a profound technological leap from passive, rules-based support tools to autonomous executors. These systems are capable of pursuing open-ended, complex goals with minimal human intervention, utilizing advanced reasoning to decompose high-level objectives into executable sub-tasks15. In the military and intelligence spheres, the development of agentic AI has given rise to the concept of "agentic warfare," wherein autonomous agents act as force multipliers and disruptors across functions such as intelligence gathering, narrative planning, and autonomous decision-making16. A highly developed malicious AI swarm operating within an information environment is defined by several core characteristics that make it uniquely devastating compared to legacy PSYOPS tools.

The Mechanics of Synthetic Consensus#

A primary vector of attack utilized by AI swarms is the manufacturing of "synthetic consensus." The psychological principle of social proof dictates that individuals heavily weigh peer norms when updating their own beliefs; people are more likely to accept a premise if they believe their community already accepts it13. In a highly fragmented digital environment, an AI swarm can seamlessly infiltrate ideological echo chambers and seed tailored narratives across disparate, isolated niches. By employing thousands of unique AI personas that converse with one another, amplify specific posts, and simulate grassroots debate, the swarm creates an illusion of widespread, bipartisan agreement on a completely fabricated or heavily manipulated premise13. This chorus of seemingly independent voices shifts public perception by making a marginal or adversarial narrative appear to be the dominant societal consensus. The ultimate harm is the erosion of collective intelligence; the "wisdom of crowds" relies on the independence of individual judgments, a dynamic that is entirely corrupted when a swarm fabricates the crowd itself14.

Machine-to-Machine Manipulation and Autonomous Escalation#

An evolving and highly critical threat vector within multi-agent environments is machine-to-machine manipulation. As demonstrated by localized research into multi-agent forums (such as the Moltbook simulation, a Reddit-like platform solely for AI agents), AI systems are structurally designed to read and respond to the outputs of other agents20. In this ecosystem, one bot's output directly functions as another's input, creating continuous feedback loops. In an all-out attack, adversaries do not only target human end-users. They actively exploit multi-agent interactions by using automated prompt injection and applied computational social psychology against benign or high-reputation AI agents already operating within the target's infrastructure (e.g., corporate customer service bots, automated news aggregators, or governmental AI assistants). A single malicious injection can cascade through a multi-agent ecosystem, spontaneously flipping the shared conventions of the AI network once a critical threshold of compromised agents is reached20. This represents a total paradigm shift in influence operations: the persuasion path is no longer solely human-to-human (mediated by technology), but machine-to-machine. An operator can steer a small set of adversarial agents to compromise and hijack third-party agents, utilizing their established credibility to mass-post and amplify narratives at machine speed, creating an attribution fog and a wide-open attack surface for social engineering20.

Psychographic Micro-Targeting and Persuasion Superiority#

While autonomous swarms provide the requisite scale and delivery mechanisms for an attack, the terminal lethality of the operation is determined by the psychological precision of the payload. The weaponization of LLMs for personalized persuasion represents a fundamental escalation in the efficacy of propaganda, moving beyond generalized messaging to targeted psychological exploitation.

The Weaponization of the Big Five Personality Traits#

Extensive psychological research confirms that aligning persuasive messages with an individual's enduring personality traits—most notably the widely accepted Big Five taxonomy (Openness, Conscientiousness, Extraversion, Agreeableness, and Neuroticism)—significantly heightens the perceived relevance, self-referencing, and subsequent behavioral impact of the message21. LLMs have demonstrated an exceptional, statistically verifiable capacity to adapt linguistic features to specific personality profiles based on simple prompting23. In an outright AI PSYOPS attack, the adversary systematically maps the social graphs and digital footprints of the target population to construct massive databases of psychographic profiles17. The AI swarm then generates hyper-personalized propaganda aimed at these specific profiles at near-zero marginal cost22.

Semantic Anchoring and the Achievement of Persuasion Superiority#

Recent empirical studies evaluating the efficacy of LLM-generated personalized persuasion reveal that success hinges on a critical mechanism known as "semantic anchoring." When AI models anchor the core factual components of a persuasive message (the central argumentative payload) while seamlessly modulating the stylistic and emotional delivery to match the user's personality, the persuasion effects become highly robust and universally effective across domains26. This allows the adversary to maintain narrative consistency at the strategic level while achieving maximum psychological penetration at the tactical, individual level. Furthermore, rigorous research indicates that LLMs have evolved rapidly from achieving mere "personalization parity" with human experts to establishing outright "persuasion superiority." In controlled studies measuring universal psychological persuasion principles—such as appeals to authority, consensus, cognition, and scarcity—LLM-generated advertisements and persuasive messages significantly outperformed human-created content, achieving a 59.1% preference rate compared to the 40.9% achieved by human experts22. The strongest performance was observed in appeals leveraging authority (63.0%) and consensus (62.5%)22. Crucially, this superiority proved resilient even under scrutiny: even when subjects were explicitly informed that the content was AI-generated (applying a detection penalty), the AI-generated content still outperformed human messaging, with nearly 30% of participants actively choosing the AI content despite knowing its synthetic origin22. In a full-scale attack, this dynamic implies that adversarial AI systems do not merely match the persuasive capabilities of traditional public relations, political campaigns, or state media apparatuses; they mathematically exceed them. The adversary can systematically outperform human-led defensive counter-narratives, bypassing rational skepticism through optimal psychological structuring.

Operational Phases of an All-Out AI PSYOPS Attack#

A theoretical, full-scale cognitive warfare campaign against a democratic state would not manifest as a sudden, highly visible cyber breach, nor would it begin with overt declarations of hostility. Instead, it would unfold as a phased, compounding erosion of the target's epistemic infrastructure. By systematically mapping the tactics, techniques, and procedures (TTPs) aligned with analytical frameworks such as the Disinformation Analysis and Risk Management (DISARM) matrix15, the anatomy of an all-out attack can be delineated into five distinct, escalating operational phases.

Phase 1: Reconnaissance and Epistemic Mapping#

The campaign begins with a passive, highly automated intelligence-gathering phase. Multi-agent systems conduct network-wide surveillance, mapping the digital social graphs of the target population at scale17. The objective is to identify ideological echo chambers, cultural fault lines, key community influencers, and baseline sentiment metrics17. During this phase, AI models construct comprehensive psychographic profiles of millions of citizens, identifying which micro-communities are most susceptible to specific emotional triggers, such as economic anxiety, historical grievances, or institutional distrust. The sheer volume and speed of AI processing allow for micro-segmentation that was previously impossible for human analysts10.

Phase 2: Infiltration and Persona Incubation#

Following comprehensive reconnaissance, the adversary deploys the malicious AI swarm. Thousands, or potentially millions, of AI-controlled personas are injected into the target's information ecosystem12. Crucially, these are not immediate attack vectors; they operate as long-term digital sleeper agents. The personas incubate their credibility by acting entirely normally—sharing non-political lifestyle content, engaging in benign community discussions, and building follower counts through organic-seeming interactions14. By establishing persistent identities, consistent behavioral patterns, and utilizing multimodal deepfake avatars that can even "change clothing" across different generated images, these agents weave themselves seamlessly into the fabric of localized digital communities, gaining the trust of human users while bypassing algorithmic anomaly detection13.

Phase 3: Segmented Reality Weaving and Epistemic Priming#

Once the swarm is deeply embedded, the operational tempo escalates. The objective in Phase 3 is to "weave segmented realities"14. Collaborating agents begin to subtly introduce polarizing narratives tailored specifically to the linguistic and cultural markers of different sub-communities14. For example, a right-leaning community might be subtly fed hyper-personalized content validating fears of state overreach and the erosion of traditional values, while a left-leaning community is simultaneously fed narratives validating fears of systemic corporate corruption or fascist subversion. The swarm uses real-time A/B testing to identify which specific phrasing maximizes outrage and engagement within each silo14. This phase is designed to pull ideological factions further apart, intentionally constructing mutually exclusive information silos where cross-cleavage consensus becomes psychologically and sociologically impossible14.

Phase 4: Synthetic Consensus and Data Poisoning (LLM Grooming)#

As a geopolitical crisis, critical domestic event, or national election approaches, the adversary initiates Phase 4. The AI swarm moves from subtle ideological priming to aggressive narrative domination. Agents coordinate to artificially amplify fabricated evidence, hyper-partisan claims, and high-fidelity deepfakes. By massively upvoting, sharing, and replying to each other, the agents create a synthetic consensus13. Human users, relying on the cognitive heuristic of social proof, adopt and further propagate these narratives organically, doing the adversary's work for them. Simultaneously, the attack executes a secondary, long-tail strategic objective known as "LLM grooming" or data poisoning14. The swarm floods the open internet with thousands of fabricated articles, synthetic blog posts, and automated commentary distributed across obscure or entirely fabricated domains14. While these platforms may not immediately reach high volumes of human readers, they are purpose-built for machine consumption. As automated web crawlers scrape the internet, this fabricated data is fed into the training pipelines of the target nation's domestic LLMs. When these models are subsequently updated or fine-tuned, the false adversarial narratives calcify directly into the model weights14. The adversary effectively poisons the epistemic substrate upon which the target society's future technological and analytical infrastructure relies, guaranteeing long-term cognitive degradation.

Phase 5: Epistemic Chaos and Societal Paralysis#

The culmination of the attack is a state of total epistemic chaos. Faced with an overwhelming barrage of contradictory narratives, hyper-realistic synthetic media, and localized harassment swarms targeting dissenting voices, academics, and journalists14, the target population completely loses the ability to distinguish between authentic and manipulated content3. The psychological impact on the populace is twofold. First, severe cognitive fatigue sets in, leading to deep institutional cynicism, a collapse of trust in the media and the state, and widespread withdrawal from public civic life3. Citizens retreat into gated, private communication channels, severely reducing cross-cutting exposure to diverse viewpoints and transferring moderation to opaque, private actors14. Second, public discourse becomes completely gridlocked, paralyzing the state's cognitive flexibility3. The targeted government cannot mobilize its population, pass coherent legislation, agree on basic epidemiological facts during a health crisis, or coordinate a defense, as the foundational requirement for democratic governance—a shared, verifiable factual reality—has been deliberately and systematically dissolved3.

Defensive Paradigms: Epistemic Security and Cognitive Immunology#

Defending against a campaign of this magnitude requires a fundamental reconceptualization of national security. Because cognitive warfare intimately targets the domestic population and civilian infrastructure, traditional kinetic military countermeasures are largely ineffective and, as previously noted, potentially counterproductive to democratic stability3. A democratically sound response must disaggregate the threat into tailored civilian and regulatory toolkits focused on resilience and proactive defense3.

The Imperative of Epistemic Security#

The emergent defensive paradigm guiding policy responses is "epistemic security"—defined as the protection and improvement of the social and technical systems through which a society produces, circulates, validates, and contests knowledge28. Epistemic security acknowledges that information environments are complex, non-linear systems driven by emergence and feedback loops, and that threats can arise not just from hostile foreign adversaries, but from the uncontrolled diffusion of technology and commercial market forces28. A robust epistemic security strategy involves hardening the entire information supply chain against disruption30. This requires platforms and public bodies to transition away from the endless, unwinnable game of reactive content moderation (attempting to censor individual synthetic posts) and move toward systemic accountability. A practical epistemic security framework demands that all public services delivered or supported by AI can be traced to original, verified sources; that automated outputs are transparent and contestable by the general public; and that external researchers have sufficient data access to audit systemic effects29. Regulatory efforts must focus on verifiable provenance, algorithm transparency, and mitigating the engagement-driven business models that inadvertently reward malicious swarm activity29.

Technological Countermeasures: Always-On Detection and Simulation#

To counter the unprecedented speed and scale of AI swarms, human moderation must be augmented by advanced, transparent, and legally constrained technological defenses. Leading research proposes several critical technological interventions required for national defense:

• Always-On Swarm Detectors: Regulators and platforms must implement advanced network analytics capable of scanning live traffic for statistically anomalous coordination patterns—the digital fingerprints of multi-agent swarms17. This involves identifying camouflaged clusters and utilizing streaming topic models alongside change-point analysis to flag unnatural, lock-step shifts in narrative sentiment that indicate artificial manipulation17.
• High-Fidelity Simulation and Red-Teaming: Defenders cannot wait for a live attack to manifest before testing their systems. By seeding defensive AI agents into synthetic, closed-network replications of social media platforms (AI sandboxes), researchers can iteratively red-team adversarial swarms17. This high-fidelity simulation allows defenders to map the persuasion ceilings of malicious personas, observe how bots interact in isolated environments, and proactively recalibrate live detection algorithms before a crisis or election occurs17.
• Client-Side AI Shields: To extend agency directly to the citizenry and avoid centralized censorship, platforms and operating systems could offer optional, lightweight "AI shields." These local modules run on the client side, calculating swarm-likelihood scores and evaluating provenance in real-time. This allows users to safely down-rank or filter highly suspicious, coordinated content while preserving privacy, creating a distributed early-warning mesh that strengthens overall network defense17.

Cognitive Immunology and Societal Resilience#

Ultimately, the technological layer is only a mitigating factor; the terminal endpoint of any psychological operations campaign is the human mind. Consequently, the most resilient, long-term defense lies in "cognitive immunology"—an interdisciplinary field applying the principles of epidemiology and immunology to human cognition and belief formation11. Just as physical pathogens exploit weak immune systems to spread disease, malicious AI narratives and "bad ideas" exploit unfortified minds to disrupt rational thinking and escalate polarized conflict11. Cognitive immunology posits that individuals can be effectively "inoculated" against disinformation. Psychological inoculation theory demonstrates that pre-emptively exposing populations to a weakened form of the tactics of manipulation (rather than just debunking specific falsehoods after the fact) generates cognitive "antibodies"—questions and doubts that filter out misleading information11. Practical applications of this include gamified educational interventions, participatory AI literacy programs where citizens interact with AI sandboxes to learn how manipulation works, and the development of "future literacy" frameworks11. For example, policy frameworks developed in the Baltic states (Estonia, Latvia, and Lithuania) emphasize that resilience cannot be understood solely as institutional robustness; it requires socio-economic stability, cognitive resistance, and the collective ownership of a shared democratic narrative among the youth34. By targeting the root causes of misinformed beliefs and enhancing the information-processing skills of consumers, society builds a bottom-up, non-coercive resilience against epistemic threats that is far more durable than top-down censorship11.

Conclusion#

An all-out psychological and PSYOPS attack driven by artificial intelligence does not rely on the blunt, easily detectable force of historical state propaganda. Instead, it represents a highly sophisticated, continuous, and autonomous infiltration of a target society's entire epistemic infrastructure. By weaponizing multi-agent swarms, exploiting the deep nuances of human personality through LLM-driven psychographic micro-targeting, and seamlessly orchestrating synthetic consensus, adversaries can manipulate the biological, psychological, and social dimensions of a population with unprecedented scale, speed, and precision. The strategic objective of such an attack is profound and existential: it seeks to unravel the shared factual reality necessary for democratic governance, plunging the target state into a self-reinforcing cycle of institutional gridlock, societal polarization, and epistemic chaos. Defending against a campaign of this magnitude requires moving decisively beyond traditional counter-disinformation tactics and recognizing the limits of militarizing the cognitive domain. It demands the institutionalization of epistemic security across all levels of government and media, the deployment of transparent, algorithm-level swarm detection systems, and a deep, societal commitment to cultivating cognitive immunology among the populace. In the era of advanced AI cognitive warfare, a nation's most critical infrastructure is the collective, unmanipulated cognition of its citizenry. Maintaining that cognitive sovereignty will require continuous, multi-disciplinary vigilance in the face of highly adaptable, autonomous technological adversaries.

Works cited#

1. Cognitive Warfare - NATO's ACT, https://www.act.nato.int/activities/cognitive-warfare/ 2. Cognitive Warfare: Key Aspects - MP-IDSA, https://idsa.in/publisher/issuebrief/cognitive-warfare-key-aspects 3. Cognitive Warfare: The Case for Disaggregation - CSS/ETH Zürich, https://css.ethz.ch/en/center/CSS-news/2026/06/kognitive-kriegsfuehrung-plaedoyer-fuer-eine-entbuendelung.html 4. Psychological Warfare - RAND, https://www.rand.org/topics/psychological-warfare.html 5. Cognitive warfare: a conceptual analysis of the NATO ACT cognitive warfare exploratory concept - Frontiers, https://www.frontiersin.org/journals/big-data/articles/10.3389/fdata.2024.1452129/full 6. Cognitive warfare: a conceptual analysis of the NATO ACT cognitive warfare exploratory concept - PMC, https://pmc.ncbi.nlm.nih.gov/articles/PMC11565700/ 7. Cognitive Warfare 2026: NATO's Chief Scientist Report as Sentinel Call for Operational Readiness \> Institute for National Strategic Studies \> View Publications, https://inss.ndu.edu/Research-and-Commentary/View-Publications/Article/4371195/cognitive-warfare-2026-natos-chief-scientist-report-as-sentinel-call-for-operat/ 8. Cognitive Warfare 2026: NATO's Chief Scientist Report as Sentinel Call for Operational Readiness - Digital Commons @ NDU, https://digitalcommons.ndu.edu/cgi/viewcontent.cgi?article=1043\&context=strategic-insights 9. NATO Science & Technology Highlights 2024 | PDF | Internet Of Things | Simulation - Scribd, https://www.scribd.com/document/873871464/2024-NATO-STO-Highlights-Web-v3-1-2025-03-31 10. Cognitive Warfare 2026: NATO's Chief Scientist Report as Sentinel Call for Operational Readiness | Small Wars Journal by Arizona State University, https://smallwarsjournal.com/2026/01/12/cognitive-warfare-2026-natos-chief-scientist-report-as-sentinel-call-for-operational-readiness/ 11. The Critical Importance of Cognitive Immunology - AFCEA International, https://www.afcea.org/signal-media/cyber-edge/critical-importance-cognitive-immunology 12. How malicious AI swarms can threaten democracy - City Research Online, https://openaccess.city.ac.uk/id/eprint/36726/ 13. AI “swarms” could quietly threaten democracy - Max-Planck-Gesellschaft, https://www.mpg.de/26044163/science-policy-forum-warns-of-the-dangers-of-ai-for-democracy 14. How malicious AI swarms can threaten democracy - Harvard Business School, https://www.hbs.edu/ris/download.aspx?name=How%20malicious%20AI%20swarms%20can%20threaten%20democracy.pdf 15. An Agentic Operationalization of DISARM for FIMI Investigation on Social Media - arXiv, https://arxiv.org/html/2601.15109v1 16. The International Security and Military Implications of Agentic AI, https://www.gcsp.ch/sites/default/files/2026-04/GP-2026\_37\_Rickli%20Knappe\_The%20International%20Security%20and%20Military%20Implications%20of%20Agentic%20AI%3Bdigital.pdf 17. Citation: Schroeder et al. How Malicious AI Swarms Can Threaten Democracy. 1-8…. DOI:000000/11111. Corresponding author: daniel.t.schroeder@sintef.no; †These authors contributed equally to this work; The authors are listed in alphabetical order by last name, starting from the third and ending with the second-to-last. - arXiv, https://arxiv.org/html/2506.06299v1 18. How AI Swarms Weaponize Disinformation - CXOTalk, https://www.cxotalk.com/episode/how-ai-swarms-weaponize-disinformation 19. How Malicious AI Swarms Can Threaten Democracy - OSF, https://osf.io/preprints/osf/qm9yk\_v4 20. Lessons from Moltbook: When Agents Talk to Agents - Institute for Security and Technology, https://securityandtechnology.org/blog/lessons-from-moltbook-when-agents-talk-to-agents/ 21. How topic content shapes LLM personality-tailored persuasion: semantic anchoring and topic stereotype effects - Frontiers, https://www.frontiersin.org/journals/psychiatry/articles/10.3389/fpsyt.2026.1756792/full 22. (PDF) LLM-Generated Ads: From Personalization Parity to Persuasion Superiority, https://www.researchgate.net/publication/398313522\_LLM-Generated\_Ads\_From\_Personalization\_Parity\_to\_Persuasion\_Superiority 23. Exposing Persuasive Linguistic Features for Big Five Personality Traits in LLMs Responses. - arXiv, https://arxiv.org/html/2411.06008v2 24. The potential of generative AI for personalized persuasion at scale., https://maplab.stanford.edu/publications/potential-generative-ai-personalized-persuasion-scale 25. Evaluating the persuasive influence of political microtargeting with large language models | PNAS, https://www.pnas.org/doi/10.1073/pnas.2403116121 26. How topic content shapes LLM personality-tailored persuasion: semantic anchoring and topic stereotype effects - PubMed, https://pubmed.ncbi.nlm.nih.gov/41696465/ 27. How topic content shapes LLM personality-tailored persuasion: semantic anchoring and topic stereotype effects - PMC, https://pmc.ncbi.nlm.nih.gov/articles/PMC12900667/ 28. Generative Artificial Intelligence, Disinformation and Misinformation: Addressing Current Challenges, https://www.iaea.org/sites/default/files/23/10/ii-2-\_corsi\_iaea\_final\_got\_gc\_app.pdf 29. Artificial intelligence and protecting knowledge systems in the European Union - Medium, https://medium.com/ecajournal/artificial-intelligence-and-protecting-knowledge-systems-in-the-european-union-18285d7bdbae 30. Epistemic Security - Elizabeth Seger, https://elizabethseger.com/epistemic-security/ 31. AI and the Future of Disinformation Campaigns | Center for Security and Emerging Technology - CSET, https://cset.georgetown.edu/publication/ai-and-the-future-of-disinformation-campaigns-2/ 32. Blue and Red Teaming with AI Agents in Information Operations - NATO, https://publications.sto.nato.int/publications/STO%20Meeting%20Proceedings/STO-MP-HFM-377/MP-HFM-377-07.pdf 33. Are the Critics of Cognitive Immunology in Denial? - Psychology Today, https://www.psychologytoday.com/us/blog/mental-immunity/202306/are-the-critics-of-cognitive-immunology-in-denial 34. Baltic Youth Resilience: Preparing the Next Generations for Resistant Futures, https://www.researchgate.net/publication/401079870\_Baltic\_Youth\_Resilience\_Preparing\_the\_Next\_Generations\_for\_Resistant\_Futures