Non-Cloud AI Warfighters: Reducing Risk of AI in the Field

An axiom of warfighting as true in 2025 as for Sun Tzu is that it is not sufficient to simply defeat an enemy, you must destroy his capacity to wage war in the future. This often means denying or disrupting the military infrastructure and supply lines that enable action at the front. History is full of examples of action targeting military infrastructure such as the allied bombings of German factories during World War II. Disruptions of military infrastructure will continue to be a core element of military strategy, but what is included in that target list is changing. Another truism of war from Sun Tzu to today is that we want our warfighters to be equipped with the best material to effectively execute their mission as safely as possible. This is true of things like rifles, aircraft carriers, howitzers, body armor, and other traditional military hardware. It is also true of the technology that plays such a crucial role in military operations today. In a world dominated by the geostrategic competition over leadership in emerging technologies, the use of advanced technologies such as artificial intelligence (AI) is a central part of any advanced nation’s warfighting capability. The recognition of AI’s utility in warfighting is hardly new, but the impact to AI from a targeted infrastructure operation requires closer attention. Building resilience to attacks on the supporting infrastructure around a critical AI system can be the difference between a timely tactical decision and disaster. Integrating non-cloud native AI into field and command center operations will build resilience and reduce attack surfaces, all without sacrificing capabilities. Time to Decision In an enemy engagement, the time it takes for commanders on the ground to synthesize and analyze the information in front of them and formulate a tactical decision can cost lives. The ability to position forces correctly, before the enemy takes his action, is critical and can be enabled by customized, specifically trained AI. Large language models (LLM) can synthesize and analyze vast quantities of data such as: Intelligence Reports Reports from Field Commanders Rules of Engagement Intercepts Enemy Capabilities Overhead Imagery The intuitive, plain human language interface allows for clear communication between the human operator and the dataset being queried, an advantage not available with non-LLM analytics. The LLM will return answers, with document citations for transparency, and commanders shorten their time to decision. The application of an LLM to the command center environment is obvious, but many of the current, cloud native LLMs present a vulnerability that field commanders cannot afford. An LLM that is connected to a cloud architecture, even if it is a secure cloud, presents an enticing piece of infrastructure for enemy attack. A cyber or physical attack against the cloud system upon which the LLM relies will cause potentially an outage or manipulation of the training data resulting in inaccurate analysis. Making timely decision on incorrect data can be disastrous for field operations and is a risk that commanders should not take. Risk Reduction, Capability Retention Instead of viewing the LLM cloud architecture as a reason not to use an LLM, there is an alternative. AI that is intended to be deployed to the field, either in a command center or with a tactical unit, should be decoupled from the cloud. Running AI locally will ensure the reduction of risk associated with AI use by eliminating a large and obvious target from the enemy target list. Without a cloud architecture to attack, the AI system being used for tactical decisions will be insulated from direct cyberattack or a physical attack on a datacenter. Share Having robust AI does not mean having to sacrifice sensitive data to the cloud or expose your operations to an unnecessary risk. Across private industry, organizations that work with sensitive data and value security are moving to private, customized, fully secure LLMs. These capabilities can run locally on a laptop or on a small on-premise server. Their customizable nature allows for commanders and staff to have access AI that is trained only on their specific mission needs reducing the risk of hallucination. Custom tuning of the model will create a model that is purpose built for the military application in question creating a robust and accurate AI capability without the cloud architecture risk. At Frontier Foundry , our Limni LLM is being customized across multiple sectors and government use cases. This kind of secure by design, fully customizable AI reduces risk to commanders and creates a strategic advantage in information synthesis and decision making. Resilience as a Weapon Lines of communication will always need protecting from enemy disruption. However, with the use of AI in field operations, it is possible to reduce the lines of communication that require protection by taking advanced AI out of the cloud. The value of AI is not lost on our adversaries, so denial and disruption of AI used for decision making is without doubt a part of the enemy’s operational planning. Reducing the attack surface for such a tactically and strategically impactful tool will itself become a weapon that can be employed to great effect on the battlefield. Without a specific infrastructure target, AI operating inside the field command post will function completely under the control of the command staff, all data, queries, and output never leaving the post. All decision-making data points will be kept out of the air and out of the cloud denying the enemy a means to disrupt strategic lines of communication. On premise AI for warfighters in the field can be decisive and the resilience it creates can itself be a weapon. Sun Tzu never had to deal with or even imagine the impact of a cyberattack against AI, but he did think extensively about the disruption of communications and about destroying an enemy’s ability to wage war. Those principles apply today. It is up to us to consider how those principles apply to the use of geostrategically important technologies such as AI. No one doubts how decisive AI can be in a tactical situation. What is in doubt is whether we can employ it securely so that we know we are making the best tactical decisions. After all, Sun Tzu and a commander in the field today want the same thing. They want to accomplish their mission with minimal loss and to destroy the enemy’s will and ability to continue to fight. AI helps us do that today. The key is to reduce the risk to our use of AI by bringing it out of the cloud an into the command post or onto a laptop in the field. The innovative approaches of Frontier Foundry in efficient, non-GPU dependent AI offer these capabilities today. Contact us for a private demonstration. Connect with us: Substack , LinkedIn , Bluesky , X , Website To learn more about the services we offer, please visit our product page. Subscribe now Leave a comment