Enhancing Malware Detection with Generative AI: Using Variational Autoencoders to Boost Machine Learning Classifiers' Performance

This collection of recent arXiv papers presents advances across malware detection, model privacy, GPU service isolation, supply-chain admission policy, and privacy leakage benchmarks. Key contributions: (1) VAEs to generate synthetic malware to augment training data and improve classifier metrics (dual-use risk — improves detection but can aid adversaries); (2) SIGIL, a canary-based membership-inference framework that reliably detects whether documents were included in LLM training sets (raises training-data privacy and provenance-forensics concerns); (3) MalTree, a phylogenetic-style malware‑