By Chris Finch, Defense Opinion Writer.
When Congress passed the 2026 National Defense Authorization Act late last year, among the priorities they wisely included in the legislation were provisions requiring the Defense Department (DOD) to ensure the security of artificial intelligence models and the data used to train them.
U.S. military and intelligence agencies are rapidly integrating AI into their missions, and AI is becoming an essential component of battlefield superiority and mission success. As reliance on AI grows, it is vital that adoption doesn’t outpace adequate security for its use.
This is particularly important for the DOD’s ambitions to deploy next-generation command and control initiatives and defense programs that will rely on AI to integrate and accelerate decision making among different and vastly dispersed units and sensors.
These programs will rely on the safe and trusted use of AI models, and the data to train those models, across military branches, allies and mission partners. One essential program whose success hinges on trusted AI models is the DOD’s combined Joint All-Domain Command and Control (CJADC2) initiative.
What AI provides for the network
CJADC2 is a far-flung program to connect sensors, service members and communications devices across all the military services into a single network.
By unifying these systems, CJADC2 will provide U.S. warfighters with critical informational and decision-making advantages through significant information and technology integration. The program aims to enhance overall mission effectiveness and provide a decisive advantage on the battlefield.
AI is central to CJADC2. It will allow rapid data processing and analysis, automation of repetitive tasks and accurate predictions and insights, enabling commanders to make faster and more informed decisions. AI will be indispensable for managing the complexity and speed required in contemporary military operations and ensuring that military responses to threats are coordinated and timely.
Vulnerabilities of the AI models
Given CJADC2’s role in future U.S. military operations, the AI models that are integral to its capabilities will very likely be the target of cyber-attacks from unfriendly nation-state actors. Every day, well-funded cyber operations are using AI to increase the sophistication and severity of potential cyber-attacks targeting critical systems of other countries.
CJADC2’s AI models will also be vulnerable to data from less secure and high-threat networks needed to continually train them. Many AI models are only accessible within a highly classified environment. But there can be times when these AI models benefit from accessing data that is sourced from an untrusted or high-threat environment. In these instances, CJADC2 must have confidence that AI models can safely access beneficial data across different classification boundaries without compromising security.
Securing the data
Adopting a “zero trust” approach for CJADC2 to safeguard AI systems and data from cyber threats should be a priority. Zero Trust is a strict cybersecurity model that assumes no user, device, workload or network segment is inherently trusted. The concept of zero trust applies to more than simply identity validation and network access. It considers all data and content entering and leaving an enterprise to be untrusted and requires assurance that it is safe.
Making this a reality requires using cybersecurity technologies that can extend data assurance between CJADC2’s participating U.S. military services, allies and mission partners, as well as across the different classification levels and high-threat networks. The data and information must be continuously verified.
Cross-domain security technologies can provide a secure bridge to enable this policy-enforced data exchange throughout CJADC2. They are advanced security technologies that safeguard against sophisticated and evolving cyber threats and serve as the secure interface between data sources, systems, users and powerful AI tools that are often separated by organizational or classification boundaries.
Importantly, cross domain security technologies can provide data-centric protection to enable the secure interoperability of next-generation systems like CJADC2.
Maintaining isolation between different domains and classified boundaries within CJADC2 will remain important. But the compatibility of cross-domain solutions, with data-centric security models and ongoing digital modernization efforts, will accelerate the safe use of AI, giving the U.S. a tangible advantage on the battlefield.
Chris Finch is a solutions architect with Everfox, which is based in Herndon, Virginia.
Are you a Defense Daily reader with a thought-provoking opinion on a defense issue? We want to hear from you.
- We welcome submissions of opinion articles on national security, defense spending, weapons systems and related areas.
- We welcome submissions from lawmakers, administration officials, industry representatives, military officials, academics, think tank experts, congressional candidates, international experts and others on issues important to the national defense community.
- We welcome a diverse range of opinions all along the political spectrum.
- Email: editor@DefenseOpinion.com
