The Intelligent Evolution of Open-Source Intelligence: Focusing on International Legion of Defence Intelligence of Ukraine

3.1. Social Media Intelligence Gathering and Sentiment Recognition

Social media has become a vital source of intelligence and a key battleground for public opinion in the Russia-Ukraine war. Platforms like Twitter (now X) and Telegram are flooded with frontline battle reports, soldiers’ accounts, civilian experiences, and even propaganda rumors. Artificial intelligence technology can significantly enhance the collection and analysis capabilities of OSINT personnel regarding social media intelligence. On one hand, NLP (Natural Language Processing) algorithms can automatically crawl and filter relevant posts, perform sentiment analysis and topic clustering, and grasp the overall direction of public sentiment. For instance, a study categorizing sentiment and stance across nearly 2 million Ukraine-related tweets from 2022-2023 revealed that war-related posts predominantly carried negative sentiment, with distinct pro-Ukraine and pro-Russia factions emerging [41,42]. By tracking sentiment curve shifts, AI can identify public opinion fluctuations triggered by major events (such as the sharp deterioration in sentiment during the war’s initial phase [43]) and reveal divergent stances among factions on specific issues. For the International Legion, this means AI can assist intelligence analysts in monitoring shifts in perceptions of the unit within foreign volunteer communities, among Ukrainian citizens, and even within Russian public opinion. For instance, when scandals emerge within the Legion, AI sentiment analysis might capture sharp shifts in supporter and detractor emotions, providing early warnings of potential morale or reputation issues. Conversely, machine learning algorithms can identify and track networks of suspicious accounts on social media. Pro-Russian and pro-Ukrainian bot accounts, for example, have been deployed to amplify specific narratives[44,45]. AI-driven network analysis can uncover these anomalous activity patterns, issuing timely alerts for suspected information manipulation campaigns [46,47]. During the Ukraine war, the OSINT community leveraged such tools to expose numerous fake accounts collectively flooding trending topics and spreading disinformation, effectively countering Russian information warfare attempts [47,48]. Regarding the International Legion, if hostile forces disseminate misinformation or fabricate negative news about the Legion on Telegram, AI can similarly trace back to the original publishers and propagation paths, assisting in analyzing whether such actions constitute part of psychological warfare. In summary, AI-enhanced social media OSINT empowers intelligence personnel to rapidly filter vast content volumes, quantify intelligence value, and discern public sentiment trends [49,50]. This holds significant importance for real-time monitoring of dynamics within the cognitive domain of warfare and for understanding the international battalions’ position within information warfare.

The graph shows the distribution of high-frequency words related to the Ukrainian conflict in the English context, including “Ukraine”, “Russia”, “war”, “NATO”, etc. “NATO” and so on. The keywords focus on war narratives, geopolitical alliances, and value claims.

The figure shows high-frequency words from the Ukrainian (Latin transcribed form) social media corpus, such as “Ukraina” “viina (war)” “peremoha (victory) “ “svoboda (freedom)”. The keywords reflect the nation-state narrative, the spirit of resistance and collective identity.

This chart displays key Russian-language (Latin transliteration) social media keywords such as “Rossiya,” “voina (war),” “mobilizatsiya (mobilization),” “sanktsii (sanctions),” and “spetsoperatsiya (special military operation).” These keywords reflect Russia’s official discourse, war justification, and the context of international sanctions.

Figures 3 to 5 respectively present the distribution of core high-frequency keywords within English, Ukrainian, and Russian social media contexts, revealing the focal points and value demands of different linguistic communities in conflict narratives.

In the English context (Figure 3), keywords predominantly include “Ukraine,” “Russia,” “NATO,” “freedom,” and “victory,” reflecting Western public opinion’s focus on international order, democratic values, and strategic confrontation. Its narrative centers on geopolitical frameworks and liberal democratic values, emphasizing support for Ukraine and countermeasures against Russia.

Within the Ukrainian linguistic context (Figure 4), high-frequency terms such as “peremoha (victory),” “svoboda (freedom),” “narod (people),” and “maibutnie (future)” reveal the nationalism and collective identity emerging during wartime. This context emphasizes national survival, independence, and future aspirations, demonstrating the psychological mobilization and social cohesion fostered by war.

In contrast, the Russian-language context (Figure 5) centers on keywords like “mobilizatsiya (mobilization),” “sanktsii (sanctions),” and “spetsoperatsiya (special military operation),” revealing a narrative heavily reliant on official discourse and military terminology. This framing not only reinforces the stance of “war justification” but also reflects Russian society’s heightened sensitivity to external sanctions and international isolation.

Overall, the comparison of these three sets of keywords reveals distinct differences in the public opinion ecosystems of different linguistic communities: the English-speaking world emphasizes international order and values, the Ukrainian-speaking community highlights national resistance and future aspirations, while the Russian-speaking community centers on the official military framework and the sanctions context. These differences reveal the multi-layered nature of information warfare and public opinion battles, highlighting the value of AI-enhanced OSINT analysis in multilingual text comparison, sentiment evolution tracking, and discourse system deconstruction.

3.2. Image Recognition and UAV Battlefield Data Analysis

High-resolution imagery and video data occupy a central position in modern conflict OSINT—from commercial satellite imagery and battlefield photographs to drone aerial footage, all contain rich intelligence. Computer vision technologies within artificial intelligence can significantly enhance the speed and accuracy of image intelligence (IMINT) analysis. On the Ukrainian battlefield, a large number of drones capture real-time combat situations. Reports indicate Ukrainian military projects have amassed over 228 years’ worth of drone footage to train AI models in recognizing tactical patterns and target locations[51]. Such models can automatically detect critical targets like tanks, artillery, and troop concentrations within video streams, geotagging them to provide commanders with near-real-time battlefield insights[30]. Particularly in small-scale infiltration operations involving international battalions, AI analysis of drone reconnaissance footage rapidly identifies threats like enemy patrols, sentry posts, and landmines, boosting mission success rates and personnel survival. Additionally, in satellite imagery, AI performs change detection: comparing satellite images from different dates to uncover new trenches, vehicle tracks, or blast patterns. This technology was used to expose Russian atrocities in Bucha—by comparing before-and-after imagery, OSINT analysis proved bodies had been present on the streets long before Russian forces arrived, debunking Moscow’s claim of “staged scenes”[52]. Similarly, AI can verify official statements regarding the International Legion’s operational achievements. For instance, when the Legion claims to have liberated a village or destroyed enemy facilities, AI-analyzed commercial satellite imagery can verify the extent of building damage or flag changes, providing independent confirmation of combat outcomes. Beyond image interpretation, facial recognition AI has also been deployed in the Ukraine conflict: Ukrainian forces use systems like Clearview to match social media photos, confirming the identities of fallen or captured soldiers[53,54]. This technology has helped Ukrainian forces identify and contact families of fallen Russian soldiers, as well as uncover enemy personnel disguised as infiltrators [55,56]. However, its risks cannot be ignored—erroneous identifications may lead to innocent civilians facing suspicion or harm at checkpoints [57]. Therefore, in practical application, Ukrainian forces emphasize using AI recognition as an auxiliary tool, with human verification ultimately required to reduce misjudgments [58]. Overall, AI-powered image and video analysis significantly accelerates the process of extracting insights from OSINT: filtering key details from thousands of photos and identifying anomalies in lengthy videos. For frontline units like the International Legion, this translates to more timely and precise intelligence support—whether conducting pre-operation reconnaissance or post-operation assessment, AI delivers efficient assistance.

3.3. Multilingual Text Mining and Semantic Analysis

The Ukraine conflict involves information sources in multiple languages, including Ukrainian, Russian, and English. The International Legion itself comprises members from diverse nations, with related materials and discussions transcending linguistic barriers. Natural Language Processing (NLP) technologies within artificial intelligence enable OSINT to overcome language barriers, mining valuable intelligence across a broader information space. First, machine translation and multilingual search tools can uniformly translate and analyze Russian Telegram channels, Ukrainian news briefings, English tweets, and other materials to build a comprehensive corpus. For instance, OSINT analysts can use AI to translate posts from Russian military enthusiast forums to understand Russian troop movements or morale; translate local Ukrainian social media posts to obtain battlefield eyewitness accounts; monitor Western media and think tank reports to track international public opinion trends. AI can automatically filter keyword-related content and apply sentiment or semantic annotations, eliminating the inefficiency of manual line-by-line reading. Second, semantic analysis and event extraction technologies can automatically extract key elements like “who-where-what-when” from vast text volumes. For instance, after an international brigade conducts an operation, media outlets in various languages may report details. AI models can extract location, participating units, and casualties from reports and social media posts to construct event knowledge graphs, aiding intelligence personnel in cross-verifying details. Furthermore, multilingual NLP facilitates monitoring public sentiment and enemy propaganda. Russian official statements or propaganda materials often contain misinformation; AI can rapidly translate them and compare against fact databases to identify inconsistencies. For instance, in early 2022, Russia fabricated Ukrainian military atrocities to justify its invasion. Ukrainian and U.S. intelligence communities swiftly debunked these false allegations using publicly available sources [59,60]. AI can automatically cross-reference Russian official statements with OSINT evidence (e.g., imagery, reports) to expose narrative inconsistencies. Regarding international battalions, Russian media sometimes smears them as “mercenaries” or exaggerates their losses. AI can routinely track such narratives, alerting analysts to whether Russia is building momentum for an operation or discrediting the battalions, thereby inferring enemy information warfare intentions[61]. Finally, the rise of large language models (LLMs) also offers possibilities for intelligence summarization. Intelligence personnel can have AI trained in military knowledge synthesize multilingual information to generate draft reports. This proves highly practical under time constraints but requires expert review to ensure accuracy and impartiality. In summary, multilingual text mining AI expands the breadth of OSINT, liberating analysis from single-language constraints and facilitating comprehensive understanding of the cross-lingual intelligence ecosystem surrounding international legions.

3.4. Information Source Traceability and Disinformation Detection

In an open information environment, one of the major challenges facing intelligence analysis is disinformation and propaganda campaigns. Artificial intelligence tools can be used to trace information sources and identify false content, thereby enhancing the credibility and reliability of OSINT. During the Ukraine conflict, Russia extensively employed deepfake videos, fabricated audio recordings, and fake accounts to confuse public opinion[62,63]. AI can play a dual role in this context: detection and tracking. First, AI can authenticate multimedia content. For instance, in March 2022, pro-Russian supporters released a fabricated video purporting Zelenskyy to “order surrender.” Despite its low resolution, it was swiftly exposed as a low-quality deepfake [64]. With the emergence of more sophisticated deepfake videos, AI algorithms can now analyze facial movement details and audio spectra to determine whether footage was generated by GAN models [65]. Cases involving the International Legion include: In 2023, hackers impersonated former Ukrainian President Petro Poroshenko to call Legion members and manipulate them into making inflammatory statements. This so-called “leaked Zoom meeting” was actually an AI-generated deepfake hoax. Subsequent investigations by The Kyiv Independent revealed it as a Russian-orchestrated disinformation campaign, highlighting the use of AI technology. Had deepfake detection tools been deployed in real-time, they might have identified video forgery indicators (such as poor lip-syncing or unnatural backgrounds) earlier, preventing Legion members from being deceived. Additionally, AI can trace information dissemination pathways. Through social network analysis algorithms, it can pinpoint the original source account that first disseminated false information and map the subsequent network of retransmissions. Many rumors exhibit clustered propagation patterns, potentially linked to specific fake news websites or bot networks. AI can map these dissemination pathways, assisting intelligence personnel in detecting state-sponsored information warfare. For instance, AI analysis revealed that certain pro-Russian false narratives originated from anonymous Telegram channels before being amplified within hours by a fleet of Twitter bot accounts, creating an illusion of viral popularity[47]. [68]. Identifying such patterns enables platforms and researchers to swiftly ban or demote relevant accounts, slowing the spread of misinformation. For international militias, similar tactics may be used to exaggerate casualties or spread rumors of internal conflicts. AI-driven attribution analysis can expose underlying manipulation, such as whether a nation-state propaganda apparatus is systematically discrediting the militia. Furthermore, AI aids in verifying the authenticity of open-source materials. While conventional OSINT techniques like digital forensics (Exif metadata checks, error-level analysis) can detect image tampering, advances in AI image generation have heightened the difficulty of manual identification. Researchers have thus developed models to recognize AI-generated images, such as by detecting unnatural textures and distortions. In summary, AI-empowered OSINT analysts can more efficiently identify and track disinformation [69,70]. This not only safeguards the accuracy of intelligence conclusions but also thwarts adversaries’ information warfare attempts on a broader scale, ensuring the dissemination of credible and reliable information.

Figure 6. False Information Detection Confidence Matrix.

Figure 6. False Information Detection Confidence Matrix.

This chart illustrates the credibility distribution across five metrics—accuracy, bias, transparency, fact-checking, and public trust—for three information sources: official, media, and social. Shade intensity indicates credibility level (on a scale of 0–1), with matrix values representing specific scores. Overall results show official sources maintain high standards in accuracy and fact-checking, while social sources exhibit generally lower credibility.

This study reveals significant differences in multi-dimensional credibility assessments across various information sources through a credibility matrix for detecting misinformation. First, official sources exhibit an average credibility rating close to 0.80, demonstrating strong performance in accuracy (0.90) and fact-checking (0.85). This indicates their robust institutionalization and authority in countering misinformation dissemination. However, public trust (0.75) and bias (0.70) still exhibit some uncertainty, suggesting that even official information requires transparent dissemination to sustain long-term social acceptance. In contrast, media sources exhibit overall medium credibility (0.60–0.75). While maintaining strong capabilities in fact-checking (0.70) and accuracy (0.75), their scores for bias (0.60) and public trust (0.60) reveal deficiencies in narrative orientation and social trust—highly correlated with recent public skepticism toward journalistic neutrality. Most critically, social sources exhibit significantly lower overall credibility (0.35–0.50), with particularly low scores in transparency (0.45) and fact-checking (0.35). This highlights their risk of becoming primary channels for disinformation dissemination. This finding validates existing literature characterizing social platforms as “low-barrier, high-diffusion” environments.

Overall, this matrix not only provides quantitative evidence of disinformation risks from various sources but also underscores the need for stricter fact-checking mechanisms and enhanced transparency measures on social platforms in future information governance. Concurrently, media outlets and official sources should further improve narrative objectivity to prevent public acceptance of accurate information from being undermined by bias and trust deficits.

3.5. Tactical Pattern Modelling and Prediction

Beyond analyzing past events, artificial intelligence can help OSINT look ahead by modeling tactical patterns and predicting conflicts. Warfare generates vast amounts of data, including combat records, troop movements, equipment losses, weather conditions, and logistics. By training machine learning algorithms on historical data, tactical patterns of both sides can be uncovered, enabling predictions about future actions or providing decision support. This approach has already shown promise in the Ukraine conflict. For instance, analysts trained models using the timing and locations of Russian missile strikes to predict potential future strike windows, enabling advance air raid warnings. Similarly, by analyzing data on Russian ground forces’ advance routes, speeds, and supply conditions, AI models can infer the “peak” and “pause” cycles of their offensive campaigns. For special operations forces like the International Legion, models can predict optimal timing and methods for operations: if intelligence indicates enemy forces are thinly spread and communications are lax in a specific area, the model can combine historical success factors for infiltration operations (such as darkness, weather conditions, enemy rotation schedules, etc.) to suggest suitable ambush windows under similar conditions. Additionally, AI can be used for wargaming and simulation. Where specialized departments once relied on complex simulations to predict battlefield scenarios, open-source information and AI models now enable a degree of simulation. For instance, by inputting troop deployments, terrain, and equipment data, AI can simulate multiple trajectories of a localized engagement. Providing such tools to International Legion commanders allows them to compare potential risks and rewards across different scenarios—while not absolutely precise, it expands tactical thinking. A more advanced application involves decision support through multi-source data fusion: U.S. and Ukrainian tech companies are collaborating on AI systems that integrate satellite imagery, electronic intelligence, weather, and terrain data to provide real-time recommendations to frontline commanders [4,71]. Reports indicate Ukrainian forces using similar systems have increased the hit rate of frontline drone strikes from 30-50% to approximately 80% [72]. This achievement stems from AI continuously refining strike precision and target selection models based on historical strike data. Thus, AI assisting in distilling tacit knowledge about “how to fight” and applying it to combat guidance is no longer a pipe dream. Of course, prediction is not infallible; warfare involves randomness and deception, and AI models may falter due to training data biases. Furthermore, overreliance on AI decision-making carries both ethical and practical risks (detailed later). However, as part of OSINT analysis, tactical pattern modeling can at least provide intelligence personnel with data-driven hypotheses to better assess enemy situations and evaluate the effectiveness of their own operations. In future operations by the International Legion, AI capable of predicting enemy reaction patterns—such as typical responses to attacks—could enable more targeted planning. This predictive capability, combined with real-time monitoring across the aforementioned dimensions (social media, imagery, intelligence), will propel conflict intelligence from passive analysis toward proactive foresight, granting Ukraine an edge in the dynamic battlefield [73,74].