Zero Trust and Clean Data: The Twin Pillars of Mission-Driven AI

At the 2025 Leidos Supplier Innovation & Technology Symposium, Rob Linger, Leidos vice president of Information Advantage, met with Tom Suder, president of ATARC, and Kurt Steege, chief technology officer at Thundercat, to discuss bringing mission-ready artificial intelligence (AI) to life. 

During their exchange, two resonant themes emerged that will significantly impact artificial intelligence (AI) and its use for years to come: establishing a secure infrastructure to obtain accurate data results and the importance of clean and usable data.

Theme one: establishing a secure infrastructure to get accurate data results

A secure infrastructure is essential to the integrity, reliability, and sustainability of any system – particularly in environments where sensitive data, critical operations, or advanced technologies like AI are involved.

At Leidos, we believe building a secure infrastructure is best achieved by incorporating consistent, rigorous verification of users, devices, and systems interacting with AI models. It comes down to a matter of trust — or zero trust. When asked how to produce a trusted system, Linger responded, “If you need a trusted system and it needs to be secure, you build security in at the beginning, you don’t bolt it in after – it’s the same with trust,” Linger said. “You have to treat it [zero trust] the same as you do with your software and infrastructure.”

Once the foundation exists, Linger noted, “It’s vitally important to have data in a state that’s AI-ready, meaning it needs to be defined in a way that’s consumable for humans and agentic AI systems to understand and read quickly.” As an emerging technology that can act independently and dynamically, agentic AI empowers users with speed and efficiency, but if not properly secured, it also introduces threats and vulnerabilities.

Agentic AI security is an area that Linger and Leidos continue putting at the forefront. With its autonomous nature and goal-directed behavior, agentic AI can be manipulated and cause harm, particularly when given access to other AI agents.

“We think about [security] from a human-centric point of view, especially with issues such as the zero trust framework, but we also need to think about that from an AI agent point of view; so what agents need to have, what data they need, and to take it a step further – how do you secure an agent at the beginning of the chain from accessing information from three other agents that it shouldn’t have access to?” Linger mused. “It’s a big security question, but it all comes down to how you curate and manage your data,” he concluded.

Theme two: clean and usable data

Today’s volume, velocity, and variety of data are staggering. By some estimates, the world will have produced approximately 180 zettabytes by the end of 2025, making AI’s ability to wade through the data more important than ever. But not all data are created equal. As organizations continually search for ways to increase their efficiency and output, it isn’t enough to have data at the ready; the data must be ready.

For example, the Defense Logistics Agency is utilizing AI to inform its decision-making process by employing more than 55 models that are at various stages of development, testing, and implementation. Harvesting data from any model can be a daunting task, but the difficulty is exacerbated when considering the different number of vendors and tools that produce data in different formats and varieties. 

When asked how the Department of Defense (restoring to Department of War) can rethink what operational efficiency looks like with AI and automation in the mix, Linger recommended document processing and areas where large volumes of information come in different formats and forms. “Those tools [AI and automation] together allow you to streamline the workflows that are important to the mission owners and get to outcomes faster,” he asserted.

While organized data remains the goal, the journey to get there is murky due to data’s “noise.”

Irrelevant, extraneous, or erroneous information within datasets can produce “noise” or obscure patterns that limit their usefulness. As a result, noise can distort data analysis, reduce prediction accuracy, and make it harder to extract valuable insight. But the ability to cut through noise and make sense of the data can lead to increased efficiency. For the U.S. Air Force, good, clean data empowered them to use predictive maintenance to resolve issues before they become problems.

As an illustrative point, Linger recalled a project where he applied AI and ML to gain operational awareness and identify efficiencies. He went on to explain how he used data to train models, allowing them to understand which components would fail and when they would fail. They used that information to preemptively schedule resources that performed repairs, maintenance, and upgrades on the equipment. 

“If you apply your advanced machine learning algorithms to both sides of the problem, you’re not only optimizing when you do your repairs on the equipment and what types of equipment you need on hand to do the repairs, but you can also manage your human capital to make sure you’re being efficient with how you schedule your people’s time, and you have the right people at the right place at the right time,” Linger concluded.

AI and ML empowered us to go from “if it ain’t broke, don’t fix it” to “we can tell when it’s going to break, so let’s get the right people in position to fix it.”

Final thoughts

AI is no longer an experiment; it’s a force multiplier helping with everything from efficiency to execution if it can be properly secured and maintained. As new technologies like agentic AI continue to take center stage and demonstrate their capabilities, Leidos remains committed to helping organizations maximize the value of their AI investments through clean, usable data that’s trustworthy and secure. 

Leidos AI capabilities