Maude-HCS: Model Checking the Undetectability-Performance Tradeoffs of Hidden Communication Systems

This collection of recent arXiv papers (Mar 5, 2026) presents multiple novel offensive and defensive findings across network covert channels, ML/LLM security, privacy-preserving ML, confidential computing, V2X safety, and program obfuscation. Key highlights: Maude-HCS — an executable modeling framework to quantify undetectability-performance tradeoffs for hidden communication systems; “Sleeper Cell” — a practical stealth backdoor vector for tool-using LLMs using a two-stage PEFT (SFT-then-GRPO) that yields trigger-specific latent malicious behavior while preserving benign performance; ZKFL-PQ—