Within the larger context of violent conflict between humans, Clausewitz’s fog of war described the battlefield in terms of sheer human performance. However, today, with the rise of large language models (LLMs) massive neural architectures trained on an exabyte of human conversation and multimodal foundation models that effortlessly combine text, vision, audio, and beyond into a shared cognitive substrate represent a standard shift not just in practice but in the ontological character of war itself. The shift is intense: LLMs, with their ability of emergent reasoning and predictive synthesis, enable the implementation of information dominance, transforming static command hierarchies into dynamic, adaptive symphonies of operational decision making. Multimodal models take that practice even further, distilling different streams of data like satellite imagery with intercepted calls, thermal signatures with semantic analysis into hyper-realistic representations of chaos, where perception and reality become muffled. As Von Clausewitz defined the “character” of war from its “nature,” warfare shifts from brute moments of kinetic exchange to a race of cognition, where victory is bestowed not to the strongest arm but the most prescient mind. Already, this is mirrored in the dilapidation of human-centered friction: the breaks between communication are replaced with prompt, probabilistically augmented reactions; it is not far-fetched to think of a time when battles will be prevailed in silicon before boots hit the disputed soil. The consequences extend further, both democratizing a foundation of asymetric power for the underdog, while increasing the observed risks for the superpower, requiring a new approach to deterrence, in a world where algorithms dream of apocalypse.
This imperative is a blade’s incision into the body of contemporary geopolitics, compelling us to engage the instinctual empiric of the Russo-Ukrainian theater as the text of AI inculcated conflict. Forget immaterial futurism, we have evidence engraved into the traumatised earth of Donbas and electromagnetic scars loiter in jammed skies, where Ukraine’s unbelievable persistence against larger conventional capabilities demonstrates how emergent technologies can turn despair into dominance. This war, too frequently outlined as a proxy of great-power rivalry, is a test tube for the transmutations of hybrid warfare. Conventional artillery duels merged with cyber incursions and drones swarms, with AI illuminating itself not as an assistant but as the sine qua non of survival. The true incubator for military invention is not in nation-state think tanks, but in a place of attrition, where blood and code unite to create the next era of defense technology.
“Operation Spiderweb,” conducted on June 1, 2025, by Ukraine’s Security Service (SBU), exemplified a sophisticated form of asymmetric thinking. More than 100 quads elevated their use from extensions of the human pilot, to autonomous drones, creating a constellation of lethality, penetrating Russia’s periphery. What occurred was merely not an advancement towards AI assisted navigation, but rather the de-familiarization in the cognitive layering of warfare (or what could classified as warfare): pre-loaded LLMs and vision-language models, trained on datasets of terrain topology and adversarial behavior, activated pathfinding in ‘real time,’ navigated contested airspace, concealed from radar, and the probabilistic fog of electronic warfare. An audacious operation targeting total of five airbases, employing the operation from Engels to Olenya, thousands of kilometers apart from each other, differentiated this from otherwise objectively unfounded military administrative tick (and assessment of altered, or completely new, capabilities): offer a fractal geometry applied to contemporary drone thought: small swarming agents causing disruption at a large scale, when emergently executed, cannonical to time islanded in actionable micro-decision (obstacle avoidance, prioritize threat capability selection) and collectively, yet near-instantaneously contributed to, a spiderweb of different systems failure in time. In a final extrapolated thought, whether theoretically classified or not: military theorists could soon be taking cue from operant conditioning and trails of real-time LLMs or cognitive learning based on topological behavior modelling to engaged drone / swarming interactions once considered hypothetical aplicado to space (land, air, maritime). Permitting semi-autonomous capabilities of the drone systems, both swarms and individual agents and close-orbit targeting application in ground, air, and open volumes of range engagement, plausibly solutions to engage adversary threats at the out injured stage – again, some considerations as the utilization of a meters’ perception and sense of engagement of drone micro-agents was most certainly applied for this operation: grouping and re-grouping in proximity to ground infrastructure as in, surely, environmental lack of threat determining to develop drone capabilities, in its programming, in like fashion.
It is important that this occasion personifies the Promethean gift of AI: Taking fire from the gods of distance and denial allows those who are cornered to use the arteries of the invader, glimmering doctrine inversion – air superiority succumbs to aerial ubiquity and he periphery of the battlefield becomes its penetrating heart.
In the electromagnetic tempest, where Russian Krasukha-4 systems unleash the denial’s symphony by severing the GPS sinews (i.e., spoofing and broadband jamming) and C2 links, this operation reveals an extensive AI autonomy ontology, which is a shield against the havoc of contested spectra. Onboard integrated sensors—e.g., LIDAR for depth mapping, inertial measurement units for dead reckoning, edge-computing cameras, and neural nets all integrated on-board—are the sensory exoskeleton that can be left behind when the vehicle is not there. At the same time, the behavior pattern outlined in the reinforcement paradigm, in which only a part of the environment is observable, leads to the non-linear nature of reaction of all given environmental/tactical jamming scenarios being work by the agent “on-the-fly”. Backup AI targeting is not a mere addition, but a gesture of latent sovereignty: LLMs compressed to lightweight frameworks—i.e. the most simplified versions of GPT-like language models or multimodal-like transformers—carry out Bayesian inference for the re-creation of target vectors and, moreover, for the analysis of acoustics that were made robust through partial observables, which are produced from degraded data (for example, thermal blooms of lazy Su-34s or the whine of turbine engines). Basically, the hand-off here is a Zen kōan to the warfare: the opposite of connectivity is not lack but abundance, which makes the removal of human oversight from the problem the agency of algorithmic, coming from the least amount of latency (i.e. seconds vs. milliseconds) from singular connectivity to distributed cognition.
The ongoing Ukraine conflict elucidates the qualitative rupture, a quantification that has transformed “Spiderweb” from a tactical ruse into a strategic weapon as the furthest spear Ukraine has deployed, over 2,000 km beyond pre-2022 incursions, it erased up to one-third of Russia’s strategic bomber fleet (most notably removing Tu-95s and Tu-22M3s fueled at dispersed air bases), causing over $2 billion in damages to materiel alone. But depth of understanding is found not by counting ink on the ledger but by looking past from within its shadow: the scale of destruction was carried out without a human input (either on board or using tele operation) while airborne, demonstrating AI’s alchemical capacity to leverage cheap silicon into asymmetric multiplier defenses, where consumer drones (price pending in the thousands) burn multi-millon dollar crown jewels. The “longest-range” aspect of “Spiderweb” ruptured spatial deterrence, not only hastening the loss of value from Russia’s continental buffer, but also with the economic ratio of billion-dollar losses to small aggregate bottom line costs.
The Mad Scientist Initiative of the U.S. Army, a leader in speculative futurology and crowdsourced polymathic provocation since 2016, and NATO’s SFA program, a biennial oracle scanning horizons to 2040, both center on AI-enabled adversary simulation as the guiding star for multi-domain dominance. The crux of matters arises from the “contested decision environments”—OODA loops impeded by fog, friction, and falsehoods, where human brains face cognitive dissonance while tripped in excess of irrational frictions. AI simulations, leveraging GANs and agent-based modeling defined by LLMs, create virtual adversaries with remarkable realism, not mere actors, not merely scripts, but adaptable minds to changing circumstances evolve tactics in virtuo, simulating Kremlin deliberate ambiguity or Beijing the swarm logics. So deeply, this represents the meta-game of warfighting: not to pre-plan against known enemies but the unknown emergent, in turn preparing joint air, sea, cyber, and space forces that come together through simulated crucibles designed to achieve cognitive overload in warfighting preparation.
“Mapping biases” uses explainable AI (XAI) to identify cognitive shortcuts. These include confirmation loops in the adversary’s inner circle and risk aversion within the Politburo. It employs natural language processing of declassified documents and open-source intelligence to make irrational thoughts seem rational. “Exposing internal cognitive blind spots” breaks the illusion of self-deception. It uses multi-model modeling to simulate feedback loops that reveal unexamined beliefs, much like a digital oracle confronting one’s own shadows. “Modeling escalation of narrative tools” depends on LLM’s insights, which are based on key events from Sarajevo to the Suez. It identifies patterns in how memes, misinformation, and overarching narratives—such as justifying “denazification” rhetoric—emerge through different phases of conflict. It simulates escalation as a branching narrative.
Overall, this is cognitive mapping at the brink of war. It focuses on preempting threats rather than predicting them. This approach, guided by AI’s neutral perspective, cuts through the complexities of enemy intentions. It creates a manageable framework where blind spots become vulnerabilities and biases turn into exploitable weaknesses. Ultimately, it serves as a safeguard against the dangerous misunderstandings that have led to disasters throughout history. This careful effort—supported by AI’s unbiased view—encourages not just softness but keen insight, similar to how a chess grandmaster anticipates an opponent’s moves. Its goal is to reduce unintentional escalations, those unpredictable sparks that arise from misinterpreted signals, like the Cuban Missile Crisis repeating in today’s rapid communication environment. It aims to add probabilistic safety measures into command structures, connecting decision trees to options for de-escalation.
Importantly, it adds a human element to an otherwise cold analysis. In AI-driven warfare, strategic empathy reminds us of our shared humanity, demonstrating that algorithms, despite their vast knowledge, struggle without an understanding of others. Otherwise, we risk automating ourselves into a state of mutual confusion.