Trojans in Artificial Intelligence (TrojAI) Final Report
Abstract
The Intelligence Advanced Research Projects Activity (IARPA) launched the TrojAI program to confront an emerging vulnerability in modern artificial intelligence: the threat of AI Trojans. These AI trojans are malicious, hidden backdoors intentionally embedded within an AI model that can cause a system to fail in unexpected ways, or allow a malicious actor to hijack the AI model at will. This multi-year initiative helped to map out the complex nature of the threat, pioneered foundational detection methods, and identified unsolved challenges that require ongoing attention by the burgeoning AI security field. This report synthesizes the program's key findings, including methodologies for detection through weight analysis and trigger inversion, as well as approaches for mitigating Trojan risks in deployed models. Comprehensive test and evaluation results highlight detector performance, sensitivity, and the prevalence of "natural" Trojans. The report concludes with lessons learned and recommendations for advancing AI security research.
Keywords
Cite
@article{arxiv.2602.07152,
title = {Trojans in Artificial Intelligence (TrojAI) Final Report},
author = {Kristopher W. Reese and Taylor Kulp-McDowall and Michael Majurski and Tim Blattner and Derek Juba and Peter Bajcsy and Antonio Cardone and Philippe Dessauw and Alden Dima and Anthony J. Kearsley and Melinda Kleczynski and Joel Vasanth and Walid Keyrouz and Chace Ashcraft and Neil Fendley and Ted Staley and Trevor Stout and Josh Carney and Greg Canal and Will Redman and Aurora Schmidt and Cameron Hickert and William Paul and Jared Markowitz and Nathan Drenkow and David Shriver and Marissa Connor and Keltin Grimes and Marco Christiani and Hayden Moore and Jordan Widjaja and Kasimir Gabert and Uma Balakrishnan and Satyanadh Gundimada and John Jacobellis and Sandya Lakkur and Vitus Leung and Jon Roose and Casey Battaglino and Farinaz Koushanfar and Greg Fields and Xihe Gu and Yaman Jandali and Xinqiao Zhang and Tara Javidi and Akash Vartak and Tim Oates and Ben Erichson and Michael Mahoney and Rauf Izmailov and Xiangyu Zhang and Guangyu Shen and Siyuan Cheng and Shiqing Ma and XiaoFeng Wang and Haixu Tang and Di Tang and Xiaoyi Chen and Zihao Wang and Rui Zhu and Susmit Jha and Xiao Lin and Manoj Acharya and Weichao Zhou and Feisi Fu and Panagiota Kiourti and Chenyu Wang and Zijian Guo and H M Sabbir Ahmad and Wenchao Li and Chao Chen},
journal= {arXiv preprint arXiv:2602.07152},
year = {2026}
}