We investigate the foundational layers of computing to build resilient, identity-aware systems. Our work includes discovering vulnerabilities in hardware and embedded systems, developing biometrics for system-level authentication, and analyzing architectural vulnerabilities through real-world CVE disclosures.

Active Projects:

• Telecommunications hardware vulnerability analysis

• Cross-device keystroke dynamics and behavioral biometric authentication

• RTOS security assessment in critical infrastructure

Selected SAIL publications:

• A. Erwin-Martinetti, A.K. Belman: "Inter-Device User Keystroke Analysis" (IEEE SVCC 2025)

• A. Erwin-Martinetti, A.K. Belman: "RTOS Security Risks in Telecommunications Hardware" (IEEE SVCC 2025)

• A. Erwin-Martinetti, A.K. Belman, M. Stamp: "Overflow Attacks on Telecommunications Hardware" (IEEE ICCCNT 2025)

Vulnerability Disclosures:

• CVE-2025-32105: Sangoma IMG2020 Buffer Overflow

• CVE-2025-32106: Audiocodes MP-11x RCE

We investigate the unique vulnerabilities of AI and ML models, including adversarial attacks against verification systems, membership inference on LLM-generated data, and studying prompt-language leakage.

Active Projects:

• Membership inference attacks on LLM-generated training data

• Synthetic media attacks on verification systems

• Deepfake generation and detection methods

Selected SAIL Publications:

• K. Jin, A.K. Belman: "Prompt-Language Leakage from LLM-Generated Data via Membership Inference on Downstream LSTMs" (IEEE SVCC 2026)

• E. Jamdar, A.K. Belman: "SyntheticPop: Attacking Speaker Verification Systems With Synthetic VoicePops" (IEEE SVCC 2025)

We apply machine learning and AI frameworks to cybersecurity challenges, including developing ML models for DDoS classification, using multi-modal behavioral signals for continuous authentication, and building transformer-based architectures for user verification.

Active Projects:

• Transformer-based multimodal user verification

• Multi-modal adversarial activity detection

• Tiny Recursive Models (TRMs) for cybersecurity

• Meta-learning ensembles for DDoS attack classification

Selected SAIL Publications:

• J. Xiong, A.K. Belman: "Parallel Stream Transformer Based Architecture for Multimodal User Verification" (IEEE AIIoT 2026)

• J. Xiong, A.K. Belman: "Multi-Modal Adversarial Activity Detection Using Keyboard and Mouse Dynamics" (IEEE AIIoT 2025)

• A.I. Kumar, G. Ishigaki, A.K. Belman: "Enhanced Multi-Class DDoS Attack Identification Using a Meta-Learning Ensemble" (IEEE ICCCNT 2025)