By Sneha Sudhakaran
As generative artificial intelligence (AI) becomes increasingly common in cybersecurity, we need to be cautious and ensure that the widely used tools do not compromise the accuracy and integrity of our work.
While these technologies significantly improve efficiency and productivity, they also introduce a critical challenge: AI hallucination.
AI hallucination is when AI models generate plausible but incorrect, misleading or fabricated information. In digital forensics research, where evidence accuracy and integrity are essential for investigative conclusions and legal proceedings, hallucinated outputs can lead to digital artifact misinterpretation and, potentially, flawed investigative outcomes.
My research, conducted alongside a group of graduate students, focuses on understanding if generative AI models produce hallucinated responses during forensic research and how researchers can systematically validate the integrity of AI-generated outputs.
Specifically, the work examines how large language models (LLMs) interpret complex forensic artifacts, such as system logs, network traces, malware reports and memory analysis results. By analyzing these interactions, we identify scenarios in which generative AI models may infer unsupported conclusions or generate fabricated explanations that appear technically convincing but are not supported by actual evidence.
A central contribution of our research, we are developing educational frameworks and validation strategies that teach researchers and students how to critically evaluate generative AI outputs.
Instead of treating AI responses as authoritative answers, our work also promotes a methodology in which AI-generated explanations are cross-checked with forensic artifacts, verified against trusted sources and analyzed using structured reasoning. This approach emphasizes the importance of evidence-based validation when integrating AI into forensic workflows.
“My research focuses on understanding if generative AI models produce hallucinated responses during forensic research and how researchers can systematically validate the integrity of AI-generated outputs.”
Sneha Sudhakaran, Assistant Professor, department of electrical engineering and computer science
Another key aspect of my AI-related research involves effective prompt design and research literacy. Because generative AI models respond heavily to how questions are framed, poorly constructed prompts can lead to incomplete or hallucinated responses.
My work explores how investigators can craft structured prompts that guide AI systems toward more reliable outputs while minimizing ambiguity.
In addition, I emphasize the importance of research reading and verification, encouraging users to compare AI responses with peer-reviewed research, technical documentation and established forensic methodologies.
The International Conference on Cyber Warfare and Security (ICCWS) recently published work I conducted with Naresh Kshetri, a faculty member at Rochester Institute of Technology, about how forensic research requires this kind of hybrid model (human plus AI) for accurate research.
This combination of prompt engineering and critical research evaluation helps reduce the risk of blindly trusting AI-generated explanations.
Sneha Sudhakaran is an assistant professor in the Department of Electrical Engineering and Computer Science. Her expertise is in cyberdefense, with special focus on cyberforensics mobile security, application security and memory analysis.
This piece was featured in the spring 2026 edition of Florida Tech Magazine.
