AFCEA: SIGNAL
May 19, 2025
Peraton is among the industry performers supporting IARPA’s TrojAI program, helping to pioneer cutting-edge techniques to detect and remediate hidden threats in AI systems. Peraton’s involvement reflects its leadership in advancing AI safety, trust, and national defense capabilities, reinforcing its role as a mission partner dedicated to protecting critical AI infrastructure from emerging threats.
The Intelligence Advanced Research Projects Activity (IARPA) is nearing completion of its TrojAI program, a groundbreaking initiative aimed at defending artificial intelligence (AI) systems from Trojan attacks—malicious manipulations designed to trigger harmful AI behavior under specific, adversary-controlled conditions.
The program has already made significant scientific contributions, generating over 150 academic publications and providing foundational data widely adopted in AI safety research. For instance, even institutions outside the program, such as the Alan Turing Institute, have leveraged TrojAI datasets—publicized by NIST—to develop firewall-like defenses for AI models, underscoring the program’s broad influence.
TrojAI focuses on identifying and mitigating backdoors in deep neural networks, including those used in large language models, computer vision, and reinforcement learning. It addresses vulnerabilities in AI training data and model architecture that could be exploited post-deployment.
