Runtime
AI PSYOPS Incident Verification and Effect Attribution Casebook
Report summary
The transition from analog psychological operations to precision, artificial intelligence-driven cognitive warfare represents a structural shift in the global information environment. Traditional mass propaganda required extensive logistical, financial, and personnel resources to identify population
Key topics
- Runtime
- AI
- .NET
- OSINT
- Research Archive
- Strategy
- Audit
- Architecture
Research provenance
For citation, use the report title and canonical URL. Archival presence does not establish authorship or promote report statements into portfolio evidence.
This page renders the archived Markdown as safe, formatted HTML. It is background research and does not become a portfolio claim without evidence review.
Full report
On this page
The transition from analog psychological operations to precision, artificial intelligence-driven cognitive warfare represents a structural shift in the global information environment. Traditional mass propaganda required extensive logistical, financial, and personnel resources to identify populations, draft localized messaging, and disseminate materials across borders. The deployment of generative artificial intelligence, large language models (LLMs), and synthetic media has effectively collapsed the marginal cost of cognitive attacks to near zero, enabling rapid, scalable influence operations that evade traditional defense mechanisms1. The objective of these operations transcends individual deception; it deliberately targets the epistemic foundations of societies, eroding shared standards for truth, trust in democratic institutions, and the fundamental habits of judgment upon which decision-making relies3. The assessment of cognitive warfare requires a taxonomy that accounts for the "constitutive invisibility" of modern influence operations3. In these operations, the targeted individual experiences induced doubt, altered values, or shifting identity organically, unaware of external manipulation, resulting in a systemic crisis of discovery where cyber tools deliver the attack but the human meaning-making layer is the ultimate target3. To navigate this environment, analysts track operations across twelve theoretical AI PSYOPS categories: Synthetic Audio Spoofing, Generative Deepfake Video, Contextual Reassignment, Media Impersonation, LLM Grooming and Data Poisoning, Targeted Reputational Deepfakes, Synthetic Persona Networks, Automated Narrative Testing, Rapid-Response Battle Damage Fabrication, Spear-Phishing Deepfakes, Automated Cognitive Probing, and Synthetic Polling2. This report provides an exhaustive, cross-category casebook of twenty documented incidents spanning the spectrum of AI-driven psychological operations. A critical vulnerability in contemporary influence analysis is the conflation of digital metrics with psychological impact. Therefore, this analysis explicitly divorces metrics of distribution—such as reach, exposure, and attention—from cognitive and behavioral effects, including comprehension, credibility, belief change, and behavioral action7. In the context of cognitive security, digital impressions, viral sharing, and media coverage must never be treated as empirical evidence of persuasion or tangible behavior change10. Every case evaluates twenty distinct variables to isolate operational mechanics from strategic effects, strictly adhering to non-operational analysis standards.
Category I: Electoral Sabotage and Democratic Subversion
The deployment of synthetic media during sensitive electoral windows exploits the vulnerability of information moratoriums and rapid news cycles. These operations aim to induce voter paralysis, suppress turnout for specific candidates, or misdirect public consensus by injecting high-friction, unverified narratives into the electorate immediately before voting commences12. The 2023 Slovak parliamentary elections serve as a prime indicator of this threat vector. Two days prior to the election, an audio recording circulated depicting Progressive Slovakia party leader Michal Šimečka allegedly conversing with a journalist regarding election rigging and manipulating the Roma minority12. Deployed during a legally mandated election news moratorium, the artifact temporarily paralyzed media organizations, preventing immediate institutional debunking and allowing the synthetic audio to propagate unchallenged across social networks12.
| Metric | Finding |
|---|---|
| Artifact or event existence | Verified. Audio file published and amplified on social platforms12. |
| Content status | AI use was confirmed. Synthetic audio deepfake overlaid on static imagery12. |
| Coordination | Networked distribution; rapidly amplified by rival political networks and Telegram channels12. |
| Actor identity | Unattributed origin; initially amplified by extremist and pro-Russian networks13. |
| Sponsorship or direction | Unproven; suspected domestic partisan actors or aligned foreign elements13. |
| Intent | Defame target candidate, depress liberal voter turnout, and sow doubts regarding election integrity12. |
| Output | Synthetic audio deepfake file12. |
| Distribution | Facebook, Instagram, Telegram, and domestic messaging forums15. |
| Availability | Highly accessible across public social media spheres15. |
| Reach | Hundreds of thousands of potential voters within the Slovak information space12. |
| Exposure | High algorithmic impression rates recorded across multiple platforms12. |
| Attention | High. Sparked mainstream media discussions post-moratorium and dominated digital discourse12. |
| Recall | High short-term recall due to the inflammatory nature of the content immediately prior to the vote. |
| Comprehension | Target audiences correctly understood the narrative that the candidate was corrupt12. |
| Credibility | High initial credibility among predisposed partisan voters; actively disputed by digital forensics post-election15. |
| Belief or attitude | Insufficient empirical data to isolate the deepfake's unique impact on long-term political ideology separate from existing bias. |
| Intention | Unmeasured. Likely reinforced existing voting intentions among adversarial constituencies without shifting neutral intent. |
| Behavior | Absent verified behavior change. While the pro-Russian candidate won, direct causal linkage between the deepfake and altered ballot-casting behavior remains unverified16. |
| Operational outcome | Successful exploitation of the media moratorium, generating widespread narrative confusion12. |
| Strategic effect | Demonstrated the efficacy of deploying synthetic media in constrained legal windows to bypass institutional fact-checking12. |
The subsequent incident involving a synthetic voice clone of United States President Joe Biden illustrates the democratization of synthetic audio spoofing. In January 2024, thousands of New Hampshire voters received a robocall featuring an AI-generated clone of Biden's voice. The artifact utilized his known catchphrases to falsely advise constituents against voting in the primary election, suggesting their votes should be saved for the general election17. The operation resulted in unprecedented federal regulatory action and highlighted the speed at which domestic actors can leverage commercial generative tools.
| Metric | Finding |
|---|---|
| Artifact or event existence | Verified. Thousands of robocalls intercepted and logged by telecommunications networks17. |
| Content status | AI use was confirmed. Voice clone generated via commercial synthetic audio software17. |
| Coordination | Centralized distribution via a telecom provider (Lingo Telecom)18. |
| Actor identity | Steve Kramer (political consultant)17. |
| Sponsorship or direction | Independent action. Actor claimed the intent was an act of civil disobedience to spur AI regulation17. |
| Intent | Ostensibly to raise awareness regarding AI dangers; functionally to deter voter participation17. |
| Output | AI-generated voice recording delivered via automated telephony18. |
| Distribution | Direct-to-consumer phone calls spoofing a known political operative's caller ID18. |
| Availability | Directed push-communication; targets did not seek out the artifact17. |
| Reach | Thousands of targeted New Hampshire voter phone numbers18. |
| Exposure | High relative to the target list; successful call connections established17. |
| Attention | Extensive national media coverage and regulatory scrutiny18. |
| Recall | High among recipients and the general public following mass media reporting. |
| Comprehension | Clear understanding of the directive to withhold votes17. |
| Credibility | Moderate to low. Promptly flagged as anomalous by sophisticated voters, though convincing in tone17. |
| Belief or attitude | No evidence of altered democratic beliefs. |
| Intention | Unverified impact on intention to vote. |
| Behavior | Absent. Turnout remained robust, and the targeted candidate won via write-in. No validated evidence of voter suppression17. |
| Operational outcome | Tactical failure in vote suppression; tactical success in generating media spectacle17. |
| Strategic effect | Catalyzed Federal Communications Commission (FCC) regulatory penalties ($6 million fine) and accelerated legislation against AI telephony spoofing19. |
Electoral subversion extends beyond advanced Western democracies. Ahead of the January 2024 national elections in Bangladesh, a deepfake video emerged depicting independent candidate Abdullah Nahod Nigar explicitly stating her withdrawal from the electoral race22. This represents a highly specific application of identity-based deepfakes, seeking to manipulate public perception by falsely attributing actions to individuals to discourage voter participation22.
| Metric | Finding |
|---|---|
| Artifact or event existence | Verified. Video surfaced on public social media ecosystems23. |
| Content status | AI use was confirmed. Deep generative methods utilized to manipulate the candidate's likeness and speech23. |
| Coordination | Disseminated by networked partisan accounts attempting to suppress opposition momentum23. |
| Actor identity | Unattributed digital operatives23. |
| Sponsorship or direction | Inferred domestic political factions seeking to benefit the ruling disposition23. |
| Intent | Suppress voter turnout for the specific candidate by manufacturing false resignation22. |
| Output | Deepfake video file23. |
| Distribution | Facebook and regional digital sharing networks23. |
| Availability | Widely accessible in targeted electoral constituencies23. |
| Reach | Tens of thousands of regional voters23. |
| Exposure | Achieved significant algorithmic spread before fact-checker intervention23. |
| Attention | Generated confusion among the electorate and required immediate debunking by campaign staff22. |
| Recall | Moderate; quickly superseded by post-election developments. |
| Comprehension | Highly effective; the message of withdrawal was unambiguous22. |
| Credibility | High initial credibility due to the low digital literacy context of the target environment23. |
| Belief or attitude | Induced temporary belief in the candidate's capitulation among exposed voters. |
| Intention | Likely intended to dissuade supporters from visiting polling stations. |
| Behavior | Absent verified behavior change. Candidate lost by a narrow margin, but empirical attribution of vote-loss directly to the deepfake is unverified23. |
| Operational outcome | Disrupted campaign operations, forcing resources into crisis communication22. |
| Strategic effect | Normalized synthetic media as a viable tool for localized voter suppression in emerging democracies14. |
During Taiwan's 2024 presidential elections, cognitive warfare escalated to include synthetic media campaigns targeting cross-strait relations. Operations featured fabricated polling data and deepfake videos portraying current President Ching-te Lai praising the opposition Kuomintang (KMT) party25. The Taiwanese information environment, highly resilient due to continuous exposure to adversarial interference, provides a crucial baseline for observing cognitive defense mechanisms against AI influence.
| Metric | Finding |
|---|---|
| Artifact or event existence | Verified. Synthetic videos and fabricated data sets were detected in the wild25. |
| Content status | AI use was confirmed. Deepfakes utilized to manipulate the politician's statements25. |
| Coordination | Networked distribution across cross-strait information platforms25. |
| Actor identity | Unattributed, assessed by researchers to align with Chinese information operations25. |
| Sponsorship or direction | Alleged state-sponsored interference targeting Taiwanese sovereignty25. |
| Intent | Confuse Democratic Progressive Party (DPP) supporters and artificially boost KMT legitimacy25. |
| Output | Deepfake video and synthetic textual data (polls)25. |
| Distribution | Taiwanese social media platforms, Line groups, and YouTube25. |
| Availability | Readily available to the Taiwanese digital public25. |
| Reach | Broad national reach across the voting populace25. |
| Exposure | Significant impressions, actively contested by Taiwanese civil society25. |
| Attention | High. Triggered national security alerts and rapid counter-messaging25. |
| Recall | High, contextualized as part of broader external interference efforts. |
| Comprehension | The fabricated endorsement was clearly understood25. |
| Credibility | Low to moderate. Taiwan's highly resilient cognitive defense ecosystem quickly identified anomalies25. |
| Belief or attitude | No verified shift in fundamental cross-strait ideological attitudes. |
| Intention | Attempted to shift voter intentions toward the opposition. |
| Behavior | Absent. The targeted candidate ultimately secured the presidency. No evidence of widespread behavior change25. |
| Operational outcome | Failed to achieve the desired electoral disruption25. |
| Strategic effect | Accelerated Taiwan's investment in AI literacy and rapid-response verification networks. |
Returning to the 2023 Slovak election, a precursor to the Šimečka incident occurred one week prior to voting. The extremist media outlet Kulturblog released an audio deepfake falsely portraying then-President Zuzana Čaputová endorsing Milan Mazurek, a candidate from the far-right Republic Movement13. This operation highlighted the utility of manufacturing endorsements from highly trusted institutional figures to legitimize fringe political movements.
| Metric | Finding |
|---|---|
| Artifact or event existence | Verified. Audio disseminated via extremist channels13. |
| Content status | AI use was confirmed. Synthetic audio cloning13. |
| Coordination | Distributed by a specific media outlet with known personnel ties to the benefiting political party13. |
| Actor identity | Kulturblog (extremist media outlet)13. |
| Sponsorship or direction | Domestic partisan actors13. |
| Intent | Legitimize a fringe candidate by manufacturing an endorsement from a highly trusted institutional figure13. |
| Output | Synthetic audio clip13. |
| Distribution | Telegram and secondary social media platforms13. |
| Availability | High availability within closed and semi-closed partisan networks13. |
| Reach | Primarily reached right-wing ideological silos before crossing into mainstream monitoring13. |
| Exposure | Moderate impression count13. |
| Attention | Garnered attention primarily from digital forensics and election watchdogs13. |
| Recall | Moderate. Overshadowed by the subsequent Šimečka deepfake incident13. |
| Comprehension | Narrative easily understood by the audience13. |
| Credibility | Low outside of deeply entrenched echo chambers; the political misalignment was overtly suspicious13. |
| Belief or attitude | Insufficient evidence of attitude shift among mainstream voters. |
| Intention | Attempted to drive undecided voters toward the far-right. |
| Behavior | Absent verified behavior change. No measurable shift in physical voting behavior attributable to this specific file. |
| Operational outcome | Minor tactical success in energizing the extremist base, failure in mainstream persuasion13. |
| Strategic effect | Demonstrated the democratization of voice-cloning technology among low-tier partisan actors12. |
Category II: Narrative Contamination and Epistemic Poisoning
Adversarial state and proxy networks continually evolve tactics to degrade the foundations of information reliability. Rather than merely presenting false information, contemporary operations seek to permanently poison the infrastructure of truth-seeking by grooming Large Language Models and generating pervasive cloned media environments4. The Russian disinformation network known as Pravda (or Portal Kombat) engaged in a campaign to flood the internet with pro-Kremlin falsehoods with the specific intent of manipulating the training data and real-time retrieval mechanisms of major AI chatbots6. By exploiting the Retrieval-Augmented Generation (RAG) pipelines of commercial models, the network ensures that users querying geopolitical events are served state-sponsored propaganda framed as objective machine output4.
| Metric | Finding |
|---|---|
| Artifact or event existence | Verified. Security audits confirmed the proliferation of fake sites and subsequent chatbot responses6. |
| Content status | AI use was confirmed. Exploited the architecture of Large Language Models (LLMs) via RAG pipelines6. |
| Coordination | Massive automated dissemination across a network of faux-news portals6. |
| Actor identity | Pravda / Portal Kombat6. |
| Sponsorship or direction | Russian state-aligned networks6. |
| Intent | Execute "LLM grooming" to ensure AI models hallucinate or cite Russian propaganda as objective truth6. |
| Output | Altered generative text outputs from major commercial AI chatbots6. |
| Distribution | Infiltrated the data scraping and RAG pipelines of commercial AIs6. |
| Availability | Available to any global user querying affected AI models regarding geopolitical events6. |
| Reach | Potentially billions of commercial AI users globally6. |
| Exposure | High. Researchers found 10 major AI chatbots repeated these narratives 33% of the time6. |
| Attention | High alarm within the cybersecurity and AI safety communities6. |
| Recall | Irrelevant for targets, as the deception occurs at the knowledge-retrieval layer4. |
| Comprehension | Users comprehended the AI outputs as authoritative factual summaries4. |
| Credibility | Extremely high. Users implicitly trust major AI platform outputs more than random web links4. |
| Belief or attitude | Strong potential to alter beliefs due to the perceived neutrality of the AI source4. |
| Intention | Unmeasured. |
| Behavior | Absent verified behavior change, though it alters information consumption habits by terminating verification. |
| Operational outcome | Highly successful epistemic contamination of commercial AI safety boundaries6. |
| Strategic effect | Highlighted a critical vulnerability in the global epistemic infrastructure, forcing AI companies to overhaul data provenance4. |
Operating alongside data-poisoning efforts is the Doppelgänger campaign. Active since at least May 2022, the Russian-aligned network utilizes generative AI and cybersquatting to clone legitimate Western news sites, including Le Monde, The Washington Post, and Der Spiegel. These cloned sites host synthetic articles designed to undermine Western support for Ukraine26. This constitutes a profound evolution in media impersonation, moving beyond isolated fake news posts into the creation of entirely simulated media environments.
| Metric | Finding |
|---|---|
| Artifact or event existence | Verified. Extensive forensic documentation of cloned infrastructure and thousands of fabricated articles26. |
| Content status | AI use was confirmed. Generative text utilized for article creation; synthetic profiles used for amplification26. |
| Coordination | Highly synchronized infrastructure utilizing automated bot amplification on Meta and X26. |
| Actor identity | Social Design Agency (SDA) and Structura National Technologies26. |
| Sponsorship or direction | Directed by the Russian Presidential Administration28. |
| Intent | Fracture European/Transatlantic resolve, discredit Ukraine, and erode trust in Western media27. |
| Output | Cloned domains, generative articles, and synthetic social media comments26. |
| Distribution | Typosquatted URLs pushed via targeted social media advertising and bot swarms26. |
| Availability | Universally accessible on the open web to users who click the deceptive links26. |
| Reach | Transnational (targeted France, Germany, USA, Ukraine)27. |
| Exposure | Millions of cumulative impressions driven by paid advertising and automated commenting28. |
| Attention | High institutional attention; resulted in massive platform takedowns and international sanctions28. |
| Recall | Low artifact recall among general populations, high thematic recall regarding Ukraine fatigue. |
| Comprehension | Clear narrative delivery utilizing localized languages27. |
| Credibility | Temporarily high when users failed to inspect URLs; degraded immediately upon forensic exposure26. |
| Belief or attitude | Contributed to ambient informational fatigue, but isolated ideological shifts are difficult to empirically attribute32. |
| Intention | Aimed to shift policy support away from military aid27. |
| Behavior | Absent verified behavior change. Western aid continued despite the campaign27. |
| Operational outcome | Sustained operational persistence despite massive digital platform countermeasures29. |
| Strategic effect | Forced fundamental shifts in Western cyber defense, resulting in US DOJ domain seizures and EU sanctions28. |
A variation of this strategy focuses on regional influence through proxy writers. Moscow-based tech firm Structura National Technologies utilized generative AI chatbots to amplify stories written by localized proxy writers across South America, targeting regional discourse34. By pairing human "meatware" for contextual authenticity with AI software for scalable distribution, the network optimized narrative injection across the Global South.
| Metric | Finding |
|---|---|
| Artifact or event existence | Verified by U.S. State Department disclosures34. |
| Content status | AI use was confirmed. AI chatbots utilized for automated dissemination and narrative amplification34. |
| Coordination | Highly synchronized integration of human authors and AI distribution bots34. |
| Actor identity | Structura National Technologies / Social Design Agency34. |
| Sponsorship or direction | Russian Kremlin34. |
| Intent | Foster anti-U.S. and pro-Russian sentiment in the Global South34. |
| Output | High-volume generative text and social media commentary34. |
| Distribution | Major social media platforms accessible in Latin America34. |
| Availability | Available to targeted demographic segments34. |
| Reach | Transnational (Central and South America)34. |
| Exposure | Unquantified officially, but assessed as high volume based on bot activity34. |
| Attention | Garnered U.S. federal sanctions and intelligence reports30. |
| Recall | Low. The tactic relies on ambient volume rather than memorable individual artifacts. |
| Comprehension | Narratives were contextually localized for high comprehension34. |
| Credibility | High, as the content originated from seemingly organic local writers before AI amplification34. |
| Belief or attitude | Long-term ideological shift is unverified but represents the core objective34. |
| Intention | Unmeasured. |
| Behavior | Absent verified behavior change. |
| Operational outcome | Achieved sustained information dominance in targeted digital sectors prior to disruption34. |
| Strategic effect | Illustrated the hybrid model of AI operations: integrating organic local authors with automated synthetic distribution networks34. |
Similarly targeting domestic discourse, a network identified by Recorded Future analysts as "CopyCop," aligned with the Russian government, shifted its focus to the 2024 US elections. The network utilized AI to generate political content and inauthentic websites, disseminating targeted narratives through a widespread architecture of YouTube videos35.
| Metric | Finding |
|---|---|
| Artifact or event existence | Verified. Network architecture and video content logged by intelligence analysts35. |
| Content status | AI use was inferred/alleged. Identified as utilizing generative AI for content scaling35. |
| Coordination | Networked cross-platform coordination bridging fake websites to YouTube amplification35. |
| Actor identity | CopyCop network35. |
| Sponsorship or direction | Russian government-aligned35. |
| Intent | Inflame partisan tensions surrounding the 2024 U.S. elections35. |
| Output | YouTube videos and inauthentic web articles35. |
| Distribution | YouTube and cybersquatted digital platforms35. |
| Availability | Readily accessible to standard internet users searching political terms35. |
| Reach | Targeted at the U.S. electorate35. |
| Exposure | High view counts on targeted YouTube distribution nodes35. |
| Attention | Identified by commercial threat intelligence sectors35. |
| Recall | Moderate among specific algorithmic consumer silos. |
| Comprehension | Standard political narratives were easily comprehended35. |
| Credibility | Moderate; disguised as standard domestic political commentary35. |
| Belief or attitude | Reinforced prevailing domestic partisan divides. |
| Intention | Unmeasured. |
| Behavior | Absent verified behavior change. |
| Operational outcome | Successfully established a bridgehead in video-based disinformation prior to detection35. |
| Strategic effect | Demonstrated the efficacy of combining text-based generative sites with automated video pipelines to bypass single-platform moderation35. |
Storm-1516 further refined the use of fake news sites by launching a highly targeted character assassination campaign against US Vice President Kamala Harris. A site titled "KBSF-San Francisco News" featured an AI-generated video of a fabricated victim, "Alisha Brown," falsely accusing Harris of a 2011 hit-and-run6. The operation combined synthetic video, stolen medical imagery, and a fabricated institutional framework.
| Metric | Finding |
|---|---|
| Artifact or event existence | Verified. Website and video deployed and tracked by Microsoft researchers6. |
| Content status | AI use was confirmed. AI-generated testimonial video utilizing stolen medical imagery6. |
| Coordination | Networked distribution; pushed by bots and inadvertently shared by prominent politicians6. |
| Actor identity | Storm-15166. |
| Sponsorship or direction | Russian state-aligned6. |
| Intent | Character assassination of a prominent political figure ahead of the 2024 election6. |
| Output | Synthetic video and fabricated news articles6. |
| Distribution | Hosted on a cybersquatted domain (Iceland registry) and amplified via X and TikTok6. |
| Availability | Available to general digital audiences6. |
| Reach | Global, focusing on the U.S. electorate6. |
| Exposure | Millions of impressions driven by algorithmic trending hashtags (\#HitAndRunKamala)36. |
| Attention | High. Debunked rapidly by CBS News and local police departments36. |
| Recall | Moderate among political operatives; low retention among the general public post-debunking. |
| Comprehension | Narrative of a criminal cover-up was unambiguous36. |
| Credibility | Briefly moderate due to the use of a seemingly legitimate news outlet structure; collapsed upon verification36. |
| Belief or attitude | Insufficient evidence to demonstrate a persistent shift in candidate favorability ratings36. |
| Intention | Unmeasured. |
| Behavior | Absent verified behavior change. No validated electoral shift resulted from this artifact. |
| Operational outcome | Temporary narrative injection achieved, but rapidly neutralized by OSINT and media verification36. |
| Strategic effect | Reaffirmed the tactical reliance of foreign actors on fabricated domestic personas to launder disinformation into adversarial information spaces6. |
Category III: Geopolitical Escalation and Synthetic Shock
Synthetic media enables state and non-state actors to manufacture geopolitical crises out of whole cloth, projecting false threats to demoralize populations, strain diplomatic alliances, or deter public activity. These operations leverage the authority of international institutions or the visceral fear of terrorism. In early 2024, the Russian-aligned Storm-1679 network launched a campaign to disparage the International Olympic Committee (IOC) and deter spectators from the Paris Games. This included a sophisticated synthetic documentary narrated by an AI clone of actor Tom Cruise, alongside deepfake news clips warning of impending terrorism37. The operation weaponized celebrity familiarity and trusted news branding.
| Metric | Finding |
|---|---|
| Artifact or event existence | Verified. The documentary and fake news segments were archived by researchers37. |
| Content status | AI use was confirmed. Voice cloning (Tom Cruise) and deepfake news anchors (Euro News, E\! News)37. |
| Coordination | Sustained, multi-platform release synchronized to coincide with Olympic milestones37. |
| Actor identity | Storm-1679 (also linked to Operation Overload / Matryoshka)37. |
| Sponsorship or direction | Russian state-aligned influence operations37. |
| Intent | Demoralize the French public, defame the IOC, and foment public fear of violence37. |
| Output | Synthetic documentary and fabricated broadcast news clips37. |
| Distribution | Telegram, X, and various video-hosting platforms37. |
| Availability | Publicly accessible; highly optimized for algorithmic recommendation38. |
| Reach | Global, with a specific operational focus on European and American audiences37. |
| Exposure | Reached millions of impressions; inadvertently amplified by high-profile figures37. |
| Attention | Garnered significant media and government attention (VIGINUM reports)32. |
| Recall | Moderate to high among targeted digital clusters due to the novelty of the Hollywood persona38. |
| Comprehension | Narrative regarding Olympic insecurity was clearly understood37. |
| Credibility | Moderately high initially due to the sophisticated brand spoofing of trusted outlets38. |
| Belief or attitude | May have induced ambient anxiety, though widespread attitude shifts against the IOC are unproven37. |
| Intention | Designed to induce the intention to cancel travel plans37. |
| Behavior | Absent verified behavior change. The Paris Olympics experienced record attendance; the campaign failed to manifest physical deterrence37. |
| Operational outcome | Achieved narrative injection but failed to achieve kinetic or behavioral disruption37. |
| Strategic effect | Forced European intelligence agencies to permanently integrate synthetic media detection into physical event security protocols32. |
A supplementary component of the Olympic disruption effort occurred in July 2024, when Storm-1679 released a deepfake video purporting to show members of Hamas threatening to carry out terrorist attacks during the Paris Games32. This artifact represented a complex multi-polar spoof, wherein a Russian-aligned network fabricated a threat from a Middle Eastern militant group against a Western nation.
| Metric | Finding |
|---|---|
| Artifact or event existence | Verified. Video surfaced globally across social networks32. |
| Content status | AI use was confirmed. Deepfake video manipulation and synthetic audio translation32. |
| Coordination | Centralized release with decentralized bot amplification32. |
| Actor identity | Storm-167932. |
| Sponsorship or direction | Russian state-aligned32. |
| Intent | Instill terror, suppress Olympic attendance, and falsely implicate a third-party geopolitical actor32. |
| Output | Deepfake terrorist threat video32. |
| Distribution | X, Telegram, and mainstream media aggregation32. |
| Availability | Widely accessible32. |
| Reach | Global32. |
| Exposure | Millions of impressions32. |
| Attention | Severe. Hamas officials themselves were forced to publicly debunk the video36. |
| Recall | Moderate. Quickly identified as a hoax by state intelligence32. |
| Comprehension | The threat of violence was clearly understood32. |
| Credibility | Briefly high due to the ambient threat environment, but rapidly degraded upon Hamas' denial and OSINT analysis32. |
| Belief or attitude | Transitory spike in anxiety; no lasting shift in geopolitical belief architectures32. |
| Intention | Induce physical avoidance of Paris32. |
| Behavior | Absent verified behavior change. The Paris Olympics proceeded without significant attendance drops37. |
| Operational outcome | Failed to achieve behavioral disruption, though it successfully hijacked media cycles32. |
| Strategic effect | Demonstrated the vulnerability of attributing synthetic threats in multi-polar environments, muddying attribution waters32. |
Shifting focus to domestic American geopolitics, Russian operatives tracked as Storm-1516 fabricated a video featuring an AI-generated persona named "Olesya," who claimed to be a Kyiv-based troll interfering in the U.S. election on behalf of Ukraine to support Joe Biden6. This operation sought to launder anti-Ukraine narratives through pseudo-independent platforms, mirroring organic whistleblowing to degrade Western support for Kyiv36.
| Metric | Finding |
|---|---|
| Artifact or event existence | Verified. Video surfaced and was amplified by identified Russian networks6. |
| Content status | AI use was confirmed. U.S. intelligence confirmed the persona and voice were synthetic6. |
| Coordination | Networked distribution; laundered through pseudo-independent platforms to mimic organic whistleblowing6. |
| Actor identity | Storm-1516 (Russian propagandist network)6. |
| Sponsorship or direction | State-aligned Russian intelligence/propaganda apparatus36. |
| Intent | Discredit Ukraine, interfere in the U.S. election, and foster domestic American political polarization6. |
| Output | Deepfake video with synthetic audio and visual persona6. |
| Distribution | X (formerly Twitter), Telegram, and fringe video-hosting sites36. |
| Availability | Accessible globally but targeted toward English-speaking American audiences36. |
| Reach | Hundreds of thousands within politically engaged digital spheres41. |
| Exposure | High impression counts before being flagged by researchers6. |
| Attention | Garnered significant scrutiny from threat intelligence firms and U.S. agencies6. |
| Recall | Moderate among specific geopolitical analysts; low among the general public. |
| Comprehension | The narrative of Ukrainian election interference was explicitly conveyed and understood6. |
| Credibility | High within algorithmic echo chambers predisposed to anti-Ukraine sentiment; low outside those silos36. |
| Belief or attitude | Likely entrenched pre-existing confirmation bias regarding foreign corruption6. |
| Intention | Unmeasured. Intended to degrade political support for the incumbent administration. |
| Behavior | Absent verified behavior change. No documented shift in aggregate U.S. voting behavior directly linked to this artifact6. |
| Operational outcome | Successfully laundered the narrative into the U.S. information ecosystem41. |
| Strategic effect | Demonstrated the evolution of "whistleblower" fabrication utilizing untraceable synthetic personas36. |
Storm-1516 further exacerbated domestic US tensions by propagating a conspiracy theory asserting that the FBI had bugged Donald Trump’s Mar-a-Lago residence during the August 2022 search36. The campaign utilized synthetic elements, staged evidence, and coordinated bot deployment to mimic a groundswell of organic outrage.
| Metric | Finding |
|---|---|
| Artifact or event existence | Verified. Coordinated posts and fabricated evidence circulated widely36. |
| Content status | AI use was alleged. Utilized manipulated imagery and highly coordinated synthetic bot networks for amplification36. |
| Coordination | High. Seeded by purported "citizen journalists" and amplified by unaffiliated proxy networks36. |
| Actor identity | Storm-151636. |
| Sponsorship or direction | Russian state-aligned36. |
| Intent | Erode domestic trust in U.S. law enforcement and stoke partisan outrage ahead of the 2024 election36. |
| Output | Fabricated whistleblower testimonials and manipulated digital imagery36. |
| Distribution | X, YouTube, and partisan American echo chambers36. |
| Availability | Readily accessible to the American electorate36. |
| Reach | Reached millions within domestic U.S. political networks36. |
| Exposure | Significant algorithmic amplification before mitigation36. |
| Attention | High intra-network attention; moderate mainstream media debunking36. |
| Recall | High among constituencies predisposed to anti-institutional narratives36. |
| Comprehension | The narrative of government overreach was explicitly clear36. |
| Credibility | High among targeted partisan demographics; zero among broader audiences36. |
| Belief or attitude | Entrenched existing institutional distrust36. |
| Intention | Unmeasured. |
| Behavior | Absent verified behavior change. No physical action or proven electoral shift directly linked to this specific artifact36. |
| Operational outcome | Successfully penetrated domestic political discourse, forcing media fact-checking36. |
| Strategic effect | Further blurred the line between domestic hyper-partisanship and foreign interference36. |
In another operation designed to drive a wedge between Western allies, Storm-1516 generated and disseminated a fake video depicting Ukrainian soldiers burning an effigy of Donald Trump36. This artifact sought to weaponize domestic American political loyalty against international military aid commitments.
| Metric | Finding |
|---|---|
| Artifact or event existence | Verified. Video surfaced and was documented by intelligence analysts36. |
| Content status | AI use was inferred/alleged. Synthetic generation and recontextualization suspected by forensic analysts36. |
| Coordination | Distributed rapidly through right-wing American channels36. |
| Actor identity | Storm-151636. |
| Sponsorship or direction | Russian state-aligned36. |
| Intent | Diminish Western support for military aid in Ukraine following Russia's invasion36. |
| Output | Fabricated video artifact36. |
| Distribution | X, Telegram, and partisan video networks36. |
| Availability | Highly accessible to targeted political segments36. |
| Reach | Targeted specific congressional constituencies and political influencers36. |
| Exposure | Achieved viral metrics before platform suppression36. |
| Attention | Addressed by disinformation researchers and media outlets36. |
| Recall | Moderate among specific voter blocs. |
| Comprehension | Narrative of Ukrainian hostility toward a US political figure was clear36. |
| Credibility | High among audiences predisposed to skepticism regarding foreign aid36. |
| Belief or attitude | Designed to solidify anti-Ukraine sentiment among specific demographics36. |
| Intention | Unmeasured. |
| Behavior | Absent verified behavior change. Did not result in immediate legislative voting shifts36. |
| Operational outcome | Succeeded in manufacturing a polarizing cultural flashpoint36. |
| Strategic effect | Validated the utility of manufacturing highly specific, visually provocative content to exploit legislative wedge issues36. |
Category IV: Reputational Assassination and Tactical Shock
The precise targeting of individuals via synthetic media bypasses rational analysis, directly targeting the limbic system through outrage and shame42. In parallel, the integration of AI-generated battle damage into ongoing kinetic conflicts collapses the distinction between the physical and cognitive domains, achieving immediate tactical paralysis2. In November 2023, an audio deepfake circulated depicting London Mayor Sadiq Khan disparaging Remembrance weekend and prioritizing pro-Palestinian marches. Released during a period of high social tension in the UK, the artifact sought to incite physical clashes between far-right protestors and law enforcement45.
| Metric | Finding |
|---|---|
| Artifact or event existence | Verified. Audio clip spread virally across multiple platforms46. |
| Content status | AI use was confirmed. Analysis revealed unnatural cadences indicative of AI voice cloning47. |
| Coordination | Amplified organically by far-right networks; initial seeding remains murky46. |
| Actor identity | Unidentified creator; amplified by domestic UK actors47. |
| Sponsorship or direction | Unknown; likely domestic rogue actors seeking racial/political friction47. |
| Intent | Incite hostility against the Mayor and spark civil disorder ahead of public demonstrations46. |
| Output | Synthetic audio clip embedded within video templates46. |
| Distribution | TikTok, X, Facebook, and WhatsApp47. |
| Availability | Highly accessible; frequently forwarded through peer-to-peer networks47. |
| Reach | Millions of UK citizens46. |
| Exposure | Hundreds of thousands of direct engagements and shares46. |
| Attention | Severe. Resulted in police reviews, mayoral statements, and national media coverage46. |
| Recall | High, due to the emotional and controversial nature of the statements49. |
| Comprehension | Narrative of mayoral betrayal and partisan bias was easily understood47. |
| Credibility | High among far-right demographics; actively disputed by authorities47. |
| Belief or attitude | Validated pre-existing animosity; unlikely to have shifted neutral attitudes given prompt debunking49. |
| Intention | Designed to incite mobilization against the Mayor and the police49. |
| Behavior | Absent verified behavior change. While clashes with police occurred, directly attributing participation to the deepfake is confounding46. |
| Operational outcome | Created an immediate crisis-management scenario for London authorities49. |
| Strategic effect | Highlighted the legal void surrounding synthetic media; authorities determined no explicit crime was committed under existing laws45. |
Reputational assassination in conservative societies often relies on gendered synthetic media to bypass political debate entirely. In the lead-up to the 2024 general election in Bangladesh, female opposition politicians, including Rumin Farhana and Nipun Roy, were targeted with highly realistic deepfake videos placing them in bikinis or swimming pools24.
| Metric | Finding |
|---|---|
| Artifact or event existence | Verified. Media files were actively circulated and collected by researchers24. |
| Content status | AI use was confirmed. AI manipulation mapping faces to non-consensual or compromising imagery24. |
| Coordination | Decentralized, but likely orchestrated by partisan opposition networks14. |
| Actor identity | Unattributed digital operatives14. |
| Sponsorship or direction | Inferred domestic pro-government or rival partisan forces23. |
| Intent | Silence, shame, and discredit female leaders within a highly conservative cultural context22. |
| Output | Synthetic image and video files24. |
| Distribution | WhatsApp, Facebook, and regional digital platforms23. |
| Availability | Highly accessible via peer-to-peer sharing23. |
| Reach | National reach across the Bangladeshi electorate14. |
| Exposure | Widespread impressions, specifically targeting the constituencies of the victims24. |
| Attention | High. Triggered trauma, rapid public responses, and international human rights condemnation44. |
| Recall | Extremely high due to the shocking cultural transgression of the imagery44. |
| Comprehension | Target audience understood the intent to frame the women as morally corrupt44. |
| Credibility | High initial credibility. The seamlessness of the technology bypassed the digital literacy levels of many voters24. |
| Belief or attitude | Induced reputational damage and likely shifted cultural attitudes toward the candidates' viability24. |
| Intention | To force withdrawal from public life and suppress support44. |
| Behavior | Absent verified behavior change on voting. However, it contributed to an environment of severe political intimidation44. |
| Operational outcome | Highly successful in creating psychological distress and forcing defensive posturing44. |
| Strategic effect | Demonstrated that in conservative societies, deepfakes do not require political substance to achieve political neutralization; character assassination suffices22. |
In the same electoral cycle, a synthetic video of exiled Bangladeshi opposition leader Tarique Rahman was circulated, showing him urging the public not to criticize Israel’s bombardment of Gaza—a highly controversial stance in a Muslim-majority nation14. This artifact sought to sever the opposition from its core geopolitical alignment.
| Metric | Finding |
|---|---|
| Artifact or event existence | Verified. Video widely distributed during the campaign23. |
| Content status | AI use was confirmed. Deepfake video manipulation14. |
| Coordination | Networked distribution across partisan channels23. |
| Actor identity | Unattributed23. |
| Sponsorship or direction | Inferred state/pro-government actors attempting to alienate the opposition's base23. |
| Intent | Drive a wedge between the opposition party and its Muslim-majority constituency over a highly emotive geopolitical issue23. |
| Output | Deepfake video23. |
| Distribution | Facebook and domestic media networks23. |
| Availability | Widely accessible23. |
| Reach | Millions of Bangladeshi voters23. |
| Exposure | High impression rates across partisan divides23. |
| Attention | High. Required immediate organizational pushback from the opposition23. |
| Recall | High, tying the candidate to a culturally resonant global conflict23. |
| Comprehension | Narrative was clearly understood23. |
| Credibility | Moderate to high among low digital-literacy demographics23. |
| Belief or attitude | Likely diminished enthusiasm among opposition supporters who believed the artifact23. |
| Intention | Unmeasured. |
| Behavior | Absent verified behavior change. The opposition ultimately boycotted the election, making specific behavioral attribution to the video impossible23. |
| Operational outcome | Succeeded in forcing the opposition to expend resources clarifying their geopolitical stance23. |
| Strategic effect | Showcased the utility of synthesizing foreign policy positions to manipulate domestic electoral dynamics23. |
The most severe evolution of AI PSYOPS is its integration into kinetic military operations. During the "Twelve-Day War" engagements between Iran and Israel, synthetic videos depicting fabricated missile strikes on Tel Aviv and downed F-35 fighter jets circulated globally across five languages within hours of actual kinetic exchanges4. The capacity to generate synthetic battle damage assessments rapidly collapses the distinction between the physical and cognitive domains2.
| Metric | Finding |
|---|---|
| Artifact or event existence | Verified. Synthetic combat footage flooded digital platforms during the conflict4. |
| Content status | AI use was confirmed. AI-generated video and recontextualized synthetic assets2. |
| Coordination | Massive, decentralized swarm distribution2. |
| Actor identity | Unattributed, likely Iranian state proxies and decentralized sympathetic networks2. |
| Sponsorship or direction | State-aligned information warfare units2. |
| Intent | Project military supremacy, demoralize the adversary, and control the global media narrative2. |
| Output | Deepfake combat footage4. |
| Distribution | Telegram, X, TikTok, and regional news aggregators2. |
| Availability | Ubiquitous during the crisis4. |
| Reach | Global4. |
| Exposure | Hundreds of millions of impressions driven by crisis algorithms4. |
| Attention | Extreme. Mainstream media networks struggled to verify footage in real-time2. |
| Recall | High, though specific artifacts blended into the broader fog of war4. |
| Comprehension | Narrative of devastating kinetic impact was instantly understood4. |
| Credibility | Extremely high in the critical first 24 hours due to the confirmation bias inherent in crisis situations2. |
| Belief or attitude | Induced temporary panic and altered perceptions of military parity2. |
| Intention | Unmeasured. |
| Behavior | Absent verified behavior change. No data confirms military command altered kinetic strategy based solely on the deepfakes, though public anxiety spiked2. |
| Operational outcome | Highly successful disruption of the adversary's information dominance2. |
| Strategic effect | Normalized synthetic battle damage as a standard operating procedure for modern kinetic engagements2. |
Conversely, democratic states are actively adapting to this environment. Israel deployed Operation PRISONBREAK, releasing deepfake content targeted at adversaries within one hour of conducting physical strikes4. This operation signifies that Western-aligned militaries now treat AI-enabled PSYOPS as essential, legitimate components of modern warfighting4.
| Metric | Finding |
|---|---|
| Artifact or event existence | Verified through analytical reports on the conflict4. |
| Content status | AI use was confirmed. Synthetic media generation aligned with physical action4. |
| Coordination | Highly centralized and coordinated by military intelligence4. |
| Actor identity | Israeli state apparatus4. |
| Sponsorship or direction | State military execution4. |
| Intent | Compound the psychological shock of kinetic strikes and overwhelm adversary decision-making cycles2. |
| Output | Deepfake content integrated with authentic strike data4. |
| Distribution | Targeted adversary communication channels and public networks4. |
| Availability | Directed specifically at adversarial populations and command structures4. |
| Reach | Regional4. |
| Exposure | High within the targeted operational theater4. |
| Attention | Immediate cognitive disruption for the adversary2. |
| Recall | Moderate; the kinetic strike generally supersedes the digital artifact in memory4. |
| Comprehension | High. Conveys absolute intelligence and operational dominance4. |
| Credibility | High, as the synthetic media was grounded by immediate, verifiable physical explosions4. |
| Belief or attitude | Diminished adversarial morale4. |
| Intention | Induce surrender, paralysis, or capitulation4. |
| Behavior | Absent verified behavior change, but theoretically designed to slow adversarial response times. |
| Operational outcome | Successfully merged the physical and cognitive domains to achieve tactical shock2. |
| Strategic effect | Validated that democratic states treat AI-enabled PSYOPS as essential, legitimate military capabilities4. |
Strategic Synthesis and Insights
The compilation and analysis of these twenty cases yield critical second- and third-order insights regarding the trajectory of AI-enabled psychological operations. The evolution observed across these incidents dictates a fundamental reassessment of cognitive defense strategies and the mechanisms by which state and non-state actors influence the global information environment. The primary defense against analog disinformation historically relied on factual verification and media literacy. However, the data reveals that the integration of AI models has shifted adversarial strategy toward structural epistemic contamination4. The Pravda Network (Case 6\) and the deployment of Doppelgänger cloned media sites demonstrate that influence operations are increasingly targeting the architectural layer of information retrieval6. By grooming Large Language Models and poisoning training data, adversaries ensure that users querying an AI assistant regarding geopolitical events are served state-sponsored propaganda framed as objective machine output4. Fact-checking frameworks are rendered obsolete when the baseline architecture of knowledge retrieval is compromised. The target experiences the induced doubt not as a partisan attack, but as an organic discovery of facts3. Furthermore, the data consistently highlights a profound disconnect between distribution metrics and psychological efficacy, reinforcing the critical distinction between exposure and persuasion7. Operations such as the New Hampshire Biden Robocall and the Paris Olympics Hamas Threat achieved massive digital reach and generated severe media spectacles, yet they failed entirely to manifest the intended physical behavior17. Voter turnout remained robust, and Olympic attendance did not crash21. This divergence indicates that while AI can effortlessly scale exposure and hijack attention networks, genuine behavioral manipulation still encounters intense cognitive friction9. Exposure to a deepfake does not invariably translate into a structural ideological shift10. The intersection of these technologies with societal structures reveals deep contextual vulnerabilities. The targeted attacks on female politicians in Bangladesh demonstrate that AI PSYOPS are devastatingly effective when they exploit localized cultural paradigms24. In conservative societies bound by stringent modesty norms, an adversary does not need to articulate a complex political argument to neutralize an opponent; generating a synthetic image that transgresses cultural taboos achieves immediate reputational destruction22. The technological capability is universal, but the psychological payload is strictly culturally contingent. Simultaneously, the convergence of the cognitive and kinetic domains during military conflict represents an alarming escalation2. The deployment of deepfakes during the "Twelve-Day War" and the proactive use of synthetic media in Israel's Operation PRISONBREAK signify the militarization of AI generative platforms4. Artificial intelligence allows for the fabrication of battle damage within hours of a kinetic event, operating well inside an adversary's decision cycle2. The strategic effect is not long-term persuasion, but immediate tactical paralysis. If a military command cannot rapidly verify whether a missile strike has occurred, decision-making is frozen, fundamentally altering the tempo of physical warfare2. Finally, the persistence of operations like Storm-1516 and Doppelgänger, despite repeated exposure and infrastructure takedowns by Western intelligence and technology platforms, indicates a doctrinal shift among adversaries27. These threat networks do not view detection as an operational failure. Instead, they operate on a "burn-and-learn" methodology, utilizing the detection mechanisms of targeted nations to actively A/B test the efficacy and resilience of their generative models4. The continuous friction provided by defensive countermeasures actively trains adversarial algorithms, ensuring that subsequent cognitive campaigns are increasingly sophisticated, culturally attuned, and structurally integrated into the modern information environment4.
Works cited
1. O'Reilly Media \- Apple Podcasts, https://podcasts.apple.com/us/podcast/oreilly-bots-podcast-oreilly-media-podcast/id1145426486
2. When Perception Becomes the Battlefield \- Small Wars Journal, https://smallwarsjournal.com/2026/07/15/when-perception-becomes-the-battlefield/
3. Invisible by design: NATO's 2026 cognitive warfare paper and the crisis of discovery, https://complexdiscovery.com/invisible-by-design-natos-2026-cognitive-warfare-paper-and-the-crisis-of-discovery/
4. AI-Driven Information Warfare: Disinformation and Psychological Manipulation, https://bisi.org.uk/reports/ai-driven-information-warfare-disinformation-and-psychological-manipulation
5. Soft Power 'Sky Net' – The Importance of Artificial Intelligence for Information Operations, https://researchcentre.army.gov.au/library/australian-army-journal-aaj/volume-19-number-1/soft-power-sky-net-importance-artificial-intelligence-information-operations
6. Entity: Storm-1516 \- AI Incident Database, https://incidentdatabase.ai/entities/storm-1516/
7. MAKING THEORY USING HEURISTICS AND BIASES TO ANALYZE PILOTS' DECISION NOT TO VACCINATE \- ProQuest, https://search.proquest.com/openview/3ca91044d0977631bbdf80014b6646a4/1.pdf?pq-origsite=gscholar\&cbl=18750\&diss=y
8. Foundations of Effective Influence Operations: A Framework for Enhancing Army Capabilities \- RAND, https://www.rand.org/content/dam/rand/pubs/monographs/2009/RAND\_MG654.pdf
9. The Attention-Information Trade-Off \- Gwern.net, https://gwern.net/doc/economics/advertising/2026-serragarcia.pdf
10. The Immensely Inflated News Audience: Assessing Bias in Self-Reported News Exposure, https://www.researchgate.net/publication/253746032\_The\_Immensely\_Inflated\_News\_Audience\_Assessing\_Bias\_in\_Self-Reported\_News\_Exposure
11. teaching public diplomacy and the information instruments of power in a complex media environment \- State Department, https://www.state.gov/wp-content/uploads/2020/08/Teaching-Public-Diplomacy-and-the-Information-Instruments-of-Power-in-a-Complex-Media-Environment-2020.pdf
12. Slovakia as the Precursor to Deepfake-Enabled Election Interference: Lessons Learned and Pathways Forward \- Proceedings of the ICWSM Workshops, https://workshop-proceedings.icwsm.org/pdf/2024\_67.pdf
13. Incident 573: Deepfake Recordings Allegedly Influence Slovakian Election, https://incidentdatabase.ai/cite/573/
14. The 2026 Electoral Information Crisis: A Quantitative Analysis of AI-Generated Misinformation and Media Impersonation in Bangladesh \- ResearchGate, https://www.researchgate.net/publication/404659494\_The\_2026\_Electoral\_Information\_Crisis\_A\_Quantitative\_Analysis\_of\_AI-Generated\_Misinformation\_and\_Media\_Impersonation\_in\_Bangladesh
15. Trolls in Slovakian election tap AI deepfakes to spread disinformation | The Straits Times, https://www.straitstimes.com/world/europe/trolls-in-slovakian-election-tap-ai-deepfakes-to-spread-disinformation
16. Get informed on the top stories of the day in one quick scan | CBC News, https://www.cbc.ca/news/canada/morning-brief-january-18-2024-1.7086086
17. New Hampshire jury acquits consultant behind AI robocalls mimicking Biden on all charges, https://www.courthousenews.com/new-hampshire-jury-acquits-consultant-behind-ai-robocalls-mimicking-biden-on-all-charges/
18. Political consultant behind fake Biden robocalls faces $6 million fine and criminal charges, https://apnews.com/article/biden-robocalls-ai-new-hampshire-charges-fines-9e9cc63a71eb9c78b9bb0d1ec2aa6e9c
19. Federal Communications Commission FCC 24-59 1, https://docs.fcc.gov/public/attachments/FCC-24-59A1.pdf
20. Company that sent fake Biden robocalls in New Hampshire agrees to $1m fine | Technology, https://www.theguardian.com/technology/article/2024/aug/22/fake-biden-robocalls-fine-lingo-telecom
21. Political consultant behind fake Biden robocalls faces $6 million fine and criminal charges \- Newsday, https://www.newsday.com/business/Biden-robocalls-AI-new-hampshire-charges-fines-v02798
22. AI, Disinformation and the Battle for Truth: How Ghana's 2024 elections exposed the new age of political deception \- WITNESS Blog, https://blog.witness.org/2025/03/disinformation-ghana-2024-election/
23. Deepfakes, Disinformation & Rationality: A Case Study of the 2024 Bangladesh Elections \- KSPP Website CMS \- Kautilya School of Public Policy, https://cms.kspp.edu.in/public/issuebrief/issuebriefdocument\_670f6b833e231.pdf
24. Deepfakes deceive voters from India to Indonesia before elections \- The Hindu, https://www.thehindu.com/sci-tech/technology/deepfakes-deceive-voters-from-india-to-indonesia-before-elections/article67700785.ece
25. Election Trends for 2024: Possible Implications for the U.S. Presidential Election, https://thesoufancenter.org/intelbrief-2024-november-6/
26. Russian Disinformation Campaign “DoppelGänger” Unmasked: A Web of Deception, https://www.cybercom.mil/Media/News/Article/3895345/russian-disinformation-campaign-doppelgnger-unmasked-a-web-of-deception/
27. Doppelganger (disinformation campaign) \- Wikipedia, https://en.wikipedia.org/wiki/Doppelganger\_(disinformation\_campaign)
28. Justice Department Disrupts Covert Russian Government-Sponsored Foreign Malign Influence Operation Targeting Audiences in the United States and Elsewhere, https://www.justice.gov/archives/opa/pr/justice-department-disrupts-covert-russian-government-sponsored-foreign-malign-influence
29. Russia's AI Interference in 2024 US Elections, https://bisi.org.uk/reports/russias-ai-interference-in-2024-us-elections
30. Imposing Sanctions on Actors Supporting Kremlin-Directed Disinformation Efforts, https://2021-2025.state.gov/imposing-sanctions-on-actors-supporting-kremlin-directed-disinformation-efforts/
31. UK action against Russian foreign information warfare \- GOV.UK, https://www.gov.uk/government/publications/uk-action-against-russian-foreign-information-warfare/new-uk-action-against-foreign-information-warfare
32. Storm-1516: A wake-up call for Europe's cognitive defence \- EPC, European Policy Centre, https://www.epc.eu/publication/Storm-1516-A-wake-up-call-for-Europes-cognitive-defence-650d24/
33. Doppelganger hub \- EU DisinfoLab, https://www.disinfo.eu/doppelganger-hub/
34. Entity: Structura National Technologies \- AI Incident Database, https://incidentdatabase.ai/entities/structura-national-technologies/
35. Escalation of State-Sponsored Cyber Threat Activity | Latest Alerts and Advisories | NJCCIC, https://www.cyber.nj.gov/Home/Components/News/News/1376/214
36. Storm-1516 \- Wikipedia, https://en.wikipedia.org/wiki/Storm-1516
37. How Russia Disrupted the 2024 Paris Olympics | Security Insider \- Microsoft, https://www.microsoft.com/en-us/security/security-insider/threat-landscape/how-russia-is-trying-to-disrupt-the-2024-paris-olympic-games
38. Russian Deepfakes Fool Media: Storm-1679 Campaign Exposed \- Reality Defender, https://www.realitydefender.com/insights/russian-propaganda-network-impersonates-major-us-news-outlets-platforms-and-gover
39. Entity: Storm-1679 \- AI Incident Database, https://incidentdatabase.ai/entities/storm-1679/
40. 20250507\_TLP-CLEAR\_NP\_SGDSN\_VIGINUM\_Technical report\_Storm-1516.pdf, https://www.sgdsn.gouv.fr/files/files/Publications/20250507\_TLP-CLEAR\_NP\_SGDSN\_VIGINUM\_Technical%20report\_Storm-1516.pdf
41. Russian US election interference targets support for Ukraine after slow start \- Microsoft On the Issues, https://blogs.microsoft.com/on-the-issues/2024/04/17/russia-us-election-interference-deepfakes-ai/
42. White Paper on Influence in an Age of Rising Connectedness \- Public Intelligence, https://info.publicintelligence.net/SMA-InfluenceConnectedness.pdf
43. Emotionally based strategic communications as a new tool in defensive cognitive warfare \- PMC, https://pmc.ncbi.nlm.nih.gov/articles/PMC12920201/
44. When AI Becomes Abuse \- LSE International Development, https://blogs.lse.ac.uk/internationaldevelopment/2025/07/21/when-ai-becomes-abuse/
45. Let's stop deepfakes damaging our democracy | Good Law Project, https://goodlawproject.org/petition/lets-stop-deepfakes-damaging-our-democracy/
46. AI-Generated Deepfake of London Mayor Sadiq Khan Incites Social Disorder \- OECD.AI, https://oecd.ai/en/incidents/2024-02-13-a372
47. No evidence Sadiq Khan 'Remembrance weekend' audio clip is real \- Full Fact, https://fullfact.org/online/remembrance-day-audio-sadiq-khan-fake/
48. Sadiq Khan AI deepfake: Explained \#shorts \#news \- YouTube, https://www.youtube.com/shorts/dLIwmaHTiCQ
49. Deepfake audio of Sadiq Khan could have caused serious disorder \- CARE, https://care.org.uk/news/2024/02/deepfake-audio-of-sadiq-khan-could-have-caused-serious-disorder
50. London Mayor Sadiq Khan says fake AI audio of him nearly led to serious disorder | BBC News \- YouTube, https://www.youtube.com/watch?v=PujHvnbRJKg
51. London Mayor Sadiq Khan confronts deepfake controversy amid AI misuse, https://dig.watch/updates/london-mayor-sadiq-khan-confronts-deepfake-controversy-amid-ai-misuse
52. Pakistan's Electoral Laboratory Shows the Drawbacks and Incentives of AI Technology for Politics \- VOA, https://www.voanews.com/a/7481049.html
53. The Burn and Learn Theory of Agency Clusterfucks | by Michael Filimowicz | Spy Novel Research | Medium, https://medium.com/spy-novel-research/the-burn-and-learn-theory-of-agency-clusterfucks-b1b51aca5bb7