Just before dawn, under a new moon, two individuals in military gear engage in a struggle within the razor wire confines of the Blue Grass Army Depot in Richmond, Kentucky. Their movements are cautious yet deliberate, attempting to gain an advantage under the dim floodlights surrounding the perimeter. A distant sentry tries to assess the situation, contemplating whether to alert his supervisor about the odd activity, but hesitates, fearing a mistaken report. Meanwhile, the struggle is immediately analyzed by Scylla, an artificial intelligence system monitoring the site from miles away. Scylla recognizes the unfolding conflict and identifies the participants—a military police officer and a known intruder—sending a rapid report to the watch captain regarding the situation. This scenario, outlined by Drew Walter, the deputy assistant secretary of defense for nuclear matters, showcases AI's crucial role in enhancing physical security. Last month, the Defense Department evaluated Scylla's capabilities at the depot, which Walter calls a significant advancement in protecting critical assets. He highlights its ability to learn in real time and reduce unnecessary alerts, addressing long-standing challenges in security. The Defense Department aims to improve surveillance and threat detection through AI, aligning with a broader strategy to integrate data, analytics, and AI across its operations. The Physical Security Enterprise and Analysis Group (PSEAG), responsible for testing Scylla, is crucial to safeguarding the nation's nuclear capabilities. Walter notes that PSEAG has unified efforts to meet national security requirements effectively. In a speech, Deputy Defense Secretary Kathleen Hicks stressed the need for quick and responsible AI integration, emphasizing its role in enhancing decision-making in defense strategies.

Chris Willoughby, the electronic security systems manager at the depot, is working to realize this vision by training Scylla’s AI software to detect and classify threats using real-time video surveillance. Scylla's demonstration showed impressive capabilities, detecting intruders and abnormal behaviors with over 96% accuracy, which greatly minimized false alarms. Walter underscored that Scylla optimizes current security systems, allowing human operators to focus on final threat responses. Scylla's potential applications extend to improving the security of the Department of Defense's strategic nuclear assets and will soon be assessed in different weather and maritime conditions at Joint Base Charleston. PSEAG is particularly interested in the algorithm's ability to counter emerging threats from unmanned systems that cross various operational domains. Despite challenges from new threats and environments, Walter and Hicks assert that AI is vital for maintaining U. S. technological superiority. The department has prioritized AI development for over a decade, aiming to leverage its strengths for responsible advancements. PSEAG's investment in AI aligns with the National Defense Strategy's focus on emerging technologies. Scylla's performance in Kentucky is seen as a pivotal step toward enhancing physical security and fully adopting AI. Walter concludes that by utilizing advanced technologies like Scylla, the department is not only bolstering security measures now but also paving the way for future innovations to safeguard the nation.