Leveraging AI for National Cyber Defense: Is an Impenetrable “AI Wall” Plausible?

By Dr. John N. Carbone

The exponential growth of Artificial Intelligence (AI) in America’s national defense is undeniable. Most visibly, Operation “Epic Fury” has been characterized as “America’s first AI-fueled war,” in which AI is significantly accelerating analysis and qualitative decision-making across the complex, asynchronous battlefield.

AI will also be essential to programs such as the Golden Dome missile defense system, which analyzes vast volumes of sensor data and intelligence to instantly classify threats, correlate tracks, and recommend optimal fire-control options.

AI can revolutionize how America conducts military operations and protects the homeland. While its use in recent conflicts and high-profile programs may grab headlines, AI is blazing a path to making a significant contribution to America’s national defense in cybersecurity.

Though it sounds like science fiction, it is conceivable that AI could advance to the point of forming an “AI Wall” that protects the entirety of America’s national security networks from adversarial cyber threats. But before that can (or should) happen, critical considerations must be thoroughly addressed.

Defense network cyber threats persist.

America’s national security organizations are increasingly under cyber threat. According to a recent report compiling survey results from U.S. government security and IT leaders, federal defense agencies experienced a 25% year-over-year surge in cyberattacks in 2025.

Given the rapidly evolving cybersecurity landscape, the Trump administration’s recent “Cyber Strategy for America” wisely acknowledged the need for a new, critical approach to cybersecurity. Rather than simply detecting and responding to cyber threats after the fact, this new approach seeks to deny cyber adversaries access to critical networks and data in the first place.

America’s national security apparatus is confronting a rapidly expanding, increasingly sophisticated cyber threat landscape. The U.S. must move decisively, acting faster and far more proactively to monitor its most sensitive defense and intelligence networks, detect anomalies in real time, and safeguard the critical information and operations that underpin national security.

AI as a National Cyber Sentry

AI can discover cyber threats at runtime and respond to them faster than ever before. Anthropic’s Claude code security tool has only been available since February, but it has found over 500 high-severity vulnerabilities in open-source code that survived decades of expert review and millions of hours of human analysis. A future in which intelligent systems provide national-scale cyber defenses by anticipating, monitoring, and reacting to cyber threats in real time is not improbable. In part, it’s already here.

In the near future, AI will likely be propelled by rapidly evolving quantum computing capabilities. While quantum computing is most often discussed in the context of the threat to encryption, it will, conversely, exponentially increase the speed of computation generally and, for AI-based cyber protection specifically, because of its inherent quantum-parallelized properties. Google’s D-Wave 2X 1000-qubit quantum processor achieved 100 million times the compute speed of a classical computer, and Quantinuum’s 56-qubit processor showed 30,000 times lower energy use than supercomputers.

This coming quantum age of computing will enable AI to process data and execute threat mitigation near-instantaneously and continuously. As these capabilities mature, they can advance AI and cyber protection as a mature, self-evolving capability; essentially providing a proverbial “AI Wall” with the scale to serve as a notional, impenetrable barrier against cyber threats from our adversaries.

But there are risks. AI is not a panacea. The Google DeepMind team recently cataloged concerns with Agentic AI targeting different parts of an agent’s operational architecture: perception, reasoning, memory, action, multi-agent dynamics, and even the human overseer. Certain considerations should be well understood before AI is used in many critical domains, particularly for applications at the national scale.

Important considerations to address

First, ensuring data integrity is essential. AI only remembers data; it does not learn from it. This is by far the most prevalent fallacy across the AI revolution. AI software executes commands assigned to it by humans, based on data and predetermined criteria it remembers. Therefore, it is critical that humans fully appreciate the context and pedigree of the datasets used to train AI factories.

All too often, humans take for granted that someone else has properly curated training data and ensured that well-known AI systems are trained on the highest-quality, fully cited, and adjudicated data. Even seasoned experts and academics can fall into this data quality trap. Every data parcel used to train an AI factory should be qualitatively curated, and its dependencies well-known to ensure relevance to a specific task. The consequences of using bad data can cascade dangerously, especially in AI processing at quantum compute speeds and scale.

Next, a key approach to stave off critical failures is to measure “trust” during the design of any new AI-based system. Trust, in this sense, is humans’ confidence in a machine performing an automated task for them. This trust can be measured by evaluating it across an Autonomic Information Continuum, which reveals how higher levels of design automation increase complexity and ambiguity while reducing trust.

Assume, for example, an engineer desires AI to automate the process of an unmanned aerial vehicle (UAV) taxiing from its hangar to a runway for takeoff. Every step, from powering up the vehicle to arriving at a designated location, requires a certain level of trust that the UAV can autonomously and as expected execute each step. Similarly, ensuring the trustworthiness of AI automation within a national-scale cybersecurity system is essential and should always be regarded as a critical requirement for deployment.

Finally, it is imperative to remember that AI is simply software code. It suffers from the same versioning and potential malware issues as other software, and the more complex the AI and its data dependencies are, the more ambiguous and vulnerable it becomes to manipulation. This is the ultimate irony in consideration of using AI to provide a comprehensive cybersecurity “Wall” for all-inclusive national security. Because vulnerabilities can be injected into an AI factory, either accidentally or maliciously, it is preferable to ensure the constitution of AI factories with watchful data integrity applied at each accessible policy enforcement point, sphere of activity, or network boundary before deployment.

Enforcement policies should automate what data moves where, in which direction, and under what conditions, while redacting what is not approved. This can be achieved through controlled, embedded, hardware-enforced boundary-isolation interfaces known as Hardsec, designed to provide comprehensive byte-by-byte data integrity and guard against adversarial inputs.

Closing thoughts

Advances in computing have the potential to dramatically transform how America achieves national security and secures our vast defense industrial base, data, and networks against ever-evolving and escalating cyber threats. But rushing headstrong into such a capability has its risks. Philosophers, authors, and educators have emphasized, “Just because we can, doesn’t mean we should,” at least not without first pursuing important safeguards.

Dr. John N. Carbone is Senior Technical Director and Chief Solutions Architect with Everfox, based in Herndon, Virginia.