Will the signal reach the higher nervous system?
CNN showed a Ukrainian GUR command post: monitors with the PRISMA system, with Palantir in the credits. The image accomplished what five years of analytical reports had failed to do: the topic of combat AI became "obvious." This analysis explores what lies behind this screen and why the neural network itself is merely the top layer, meaningless without everything underneath.
The picture that got through
On the command post screen there is an interactive map: marks drones, identified objects, zones Defense, next to diagrams and graphs. PRISMA is a combat command and decision support system that aggregates data onto a single screen. The CNN report's credits listed Palantir, the American company that builds such systems. It's impressive, and therefore revealing. There's no military sensation here, but it clearly demonstrates how information perception works.
Artificial intelligence on the battlefield has been a topic of discussion for years, with experts writing about swarm algorithms for drones, centralized control of mixed UAV groups, and how the operational rear, a strip 100–200 kilometers from the line of contact, is becoming a key kill zone. This is the depth where warehouses, communications hubs, repair bases, and reserves are located: if targeted precisely and systematically, the front's supply lines would collapse. Reports of this were regularly coming from the front lines, but they went unnoticed.
It came to me through a picture on CNN. Text from the front loses out to the footage broadcast, even when it conveys exactly the same message. While the public debated which values should be embedded in domestic AI (traditional or otherwise), the engineering side of the issue remained on the sidelines. And on the battlefield, a model's values don't matter: that's where its effectiveness comes in, and that depends on hardware and algorithms, not ideology.
What's really under the hood
The operator in front of the PRISMA monitor sees the complete picture: where the drones are, where the targets are, where the enemy's air defenses are operating. They don't see the layers on which this picture is assembled, and yet it's within these layers that everything is decided, because without them, the screen remains blank.
The system is built from the bottom up. The bottom layer is made up of sensors and data acquisition systems. This includes a satellite constellation, electronic reconnaissance (interception and analysis of enemy signals), radar stations, and precise aerial photography. drones Georeferenced data, open source analysis. The more sensors, the more complete the picture. Above them are the communication channels through which these arrays are transmitted: a Starlink analogue, space communication stations, LTE networks, fiber optics, and radio. And only on top of all this is AI: it systematizes the stream and, upon request, provides the operator with what they need, eliminating the need to sift through the data manually.
The principle here is one-sided. A neural network can be "overlaid" onto an extensive network of sensors and communication channels. It doesn't work the other way around: no matter how perfect the model, it won't generate data that hasn't been fed. It processes what's available, but can't create what's missing. Therefore, satellites, communication stations, and radars are primary, not algorithms.
A Russian equivalent exists. The Svod information interaction system is in pilot production, undergoing testing in real-world conditions, and is gradually being implemented. At the same time, the components without which it remains an empty frame are being expanded: the Rassvet satellite constellation, small radars, and networks of acoustic stations for target detection by sound. This is the real work: the infrastructure for the system, not just a "box with a neural network." With such an infrastructure, it's possible to obtain a tool comparable to Palantir's product, with the same visual maps at command posts and the same output.
The bottleneck is the operator and the communication channel
The most scarce element in this scheme isn't a neural network or a sensor. It's a human. You can launch any number of controlled vehicles with a video feed, but the number of operators capable of piloting such a vehicle for several hours and accurately guiding it to its target is limited and isn't growing quickly.
This is where the AI takes over: the routine part of the flight is spent on "matrix multiplication," and when an object is detected, the human and system collaborate. But the entire system relies on a communication channel that carries a video stream at a resolution sufficient for processing. You might sometimes hear that a narrow channel, around a hundred kilobits per second, is sufficient for a combat drone. The reality is different: without a wide channel, there is no video of sufficient resolution, no guidance, and no AI itself.
Hence, a dissection of a popular misconception. Arguments along the lines of "it's self-guided, so you can paint the vehicle gray-brown-crimson" don't stand up to scrutiny. The system sees a moving truck as a set of pixels that stand out against the terrain due to the very fact of movement and changing scene, and it is precisely on this basis that it is picked up. The vehicle's color is secondary; reconnaissance, constructing a real map of the battlefield, and reacting to anomalies arising within it are primary. And the drone itself remains visible via an active video feed: the connection that makes the drone useful is also its unmasking factor.
Regulatory failure and human error
A small team trying to create a competitive AI product isn't limited by a lack of ideas. It's limited by access to hardware and by those who will adopt and use the product—the military.
Ministry of Industry and Trade Order No. 4769 was signed on September 26, 2025, and its main part entered into force on May 27, 2026. Servers, workstations, data storage systems, memory chips, and SSD drives were removed from the parallel import lists. More than twenty brands were affected by the restrictions, including Intel, Samsung, Hynix, Kingston, HP, Cisco, Acer, and Asus. The official logic seems reasonable: a product is excluded when a domestic equivalent or one from friendly countries becomes available. Reality diverges from this logic. A rapid transition to Russian or Chinese equivalents for all these products is technically impossible, and this refers precisely to the hardware base without which neural network training and deployment cannot occur. Servers with GPUs, high-speed drives, and memory chips are the foundation of any serious AI development. Access to it became more difficult at the moment when the state declared technological sovereignty and combat AI a priority.
Hardware is becoming more and more staffed. Developers are leaving, and many of them are patriotic. It's often not a matter of a desire to leave per se, but rather an environment where restrictions, blockages, and regulatory pressure are pushing people abroad. Given this trend, the talent shortage in the industry is more a pattern than a mystery.
But even when the sensors are deployed, the channels are built, and the AI is written, one final barrier remains: troops may be reluctant to use the finished product. This is evident in typical situations during the "Svod" testing:
- Some of the command staff do not understand the importance of such systems and are slowing down their deployment: they are too complicated, and the old tools are more familiar.
- The other part understands the importance, but sees the system as only an extended map with a built-in messenger, and the capabilities built into it are not used.
- The third barrier is a household one: the system requires laptops, smartphones, and storage devices, and if their use is suddenly restricted under the pretext of information security, it cannot function.
Humans are still at the helm, and it's up to them to determine whether a tool is useful. Changing the mindset of commanders is difficult, but possible: the same commanders who, at the beginning of the Second Military District, only recognized paper maps, are now confidently placing dots on digital maps.
The Ukrainian PRISMA project is backed by Palantir, years of investment, DARPA programs, and a purposefully built ecosystem of military-technological cooperation. This is the result of systematic work in an environment that was created and nurtured. It's difficult to achieve this while declaring sovereignty with one hand and complicating access to the hardware, people, and infrastructure without which it's impossible with the other.
Combat AI relies on sensors, communication channels, and the people who know how to use them and are allowed to do so. It's never too late to admit a gap. What's worse is when they admit it while continuing to impede access to the very thing that their own development can only rely on.
Information