Synthetic Media in the Information Battlespace: Scalable Evaluation for Detection, Attribution, and Characterization Analytics
Abstract:
As synthetic media becomes weaponized in influence operations and cognitive warfare, the ability to assess the authenticity and intent of multimodal information is essential to sustaining information integrity and decision advantage. This paper introduces a scalable multimodal evaluation framework for assessing analytics that detect, attribute, and characterize synthetic and manipulated media across text, image, audio, and video. Drawing on insights from the DARPA Semantic Forensics (SemaFor) program, the authors report results across 111 evaluation tasks and identify strengths and vulnerabilities in analytic performance. Findings highlight the need for explainable, context-aware analytic frameworks that support cognitive security and operational decision making.
AUTHORS
Aptima, Inc.
Woburn, Massachusetts United States of America
Laura Cassani serves as the Deputy Director of the Intelligent Performance Analytics Division at Aptima, Inc., where she spearheads initiatives at the forefront of Artificial Intelligence research to enhance both human and machine performance. Cassani has served as Principal Investigator on numerous high-impact research and development efforts for the Department of Defense, including projects with DARPA, the Office of Naval Research, Marine Corps Systems Command, Air Force Research Laboratory, and the Office of the Secretary of Defense. Her recent work focuses on the application of generative AI for defense and national security missions. She led as PI the test and evaluation technical area for DARPA’s Semantic Forensics (SemaFor) program, developing methodologies to assess AI algorithms that detect, attribute, and characterize semantic manipulation in synthetic multimedia. She continues to oversee the transition of SemaFor capabilities through the ULRI Digital Safety Research Institute (DSRI), building national research capacity in synthetic media detection. Cassani holds an MA in security studies from Georgetown University and a BA in international relations from Boston University.
Aptima, Inc.
Woburn, Massachusetts United States of America
Peter Bautista is an Innovation Lead in the Intelligence Performance Analytics Division at Aptima, Inc. He has a diverse skill set in the realm of data science, machine learning, and natural language processing, with a background in both academic and industry settings. His expertise spans across developing innovative solutions for large language model (LLM) applications and evaluation, visualization, and management, including the implementation of agentic workflows and reinforcement learning frameworks. Bautista received an MS in computer science from California State University, Fullerton, and a BS in physics from University of California, Riverside.
Aptima, Inc.
Woburn, Massachusetts United States of America
Michael Davinroy is an AI Engineer in the Intelligence Performance Analytics Division at Aptima, Inc. He uses his experience in machine learning, low-level systems programming, and algorithmic game theory to design, implement, and evaluate new highly efficient intelligent agents that act in uncertain environments. Davinroy received an MS in computer science from Northeastern University and a BA with honors in computer science from Swarthmore College.
Aptima, Inc.
Woburn, Massachusetts United States of America
Lauren Fortier is a Research Engineer in the Intelligent Performance Analytics Division at Aptima, Inc., specializing in the design, evaluation, and implementation of human-centered automation and decision-support tools, as well as in experimental design and statistical analysis. Her experience in machine learning and applied statistics is honed across her work in data analysis, visualization, graph networks, and reinforcement learning. Fortier received an MS in statistical practice from Boston University, and a BS in psychology from Western New England University.
Aptima, Inc.
Woburn, Massachusetts United States of America
Tatiana Toumbeva is a Senior Scientist and Team Lead in the Training, Learning, and Readiness Division at Aptima, Inc. with expertise in the areas of learning and training, leadership, selection and assessment, and organizational change. Toumbeva applies learning principles and systems design to develop adaptive training capabilities in complex instructional and operational settings. She holds a PhD in Industrial/Organizational Psychology from Bowling Green State University, a Global Professional in Human Resources (GPHR) certification from HRCI, a change management practitioner certification from Prosci, an MA in mental health from Boston University, and a BS in biology from the University of South Florida.
Aptima, Inc.
Woburn, Massachusetts United States of America
Svitlana Volkova is Chief of AI in the Office of Science and Technology at Aptima, Inc. She spearheads the company's initiatives to develop trustworthy and human-centric compound AI systems that tackle complex real-world problems for the Department of Defense and other government agencies. Volkova's research focuses on advancing natural language processing and machine learning techniques, with an emphasis on graph neural networks, causal inference, and multimodal frontier models. Her work has pioneered methods for AI-powered analytics to explain complex social systems and behaviors. She received her PhD in computer science from Johns Hopkins University, and MS degrees in computer science from Petro Mohyla Black Sea State University, Ukraine, and Kansas State University.
Published In
Journal of Information Warfare
The definitive publication for the best and latest research and analysis on information warfare, information operations, and cyber crime. Available in traditional hard copy or online.
Quick Links
Archive
