Side-Channel

LLM assistance in reverse engineering, side-channel device tracking in exclusive-use IoT, and defensive measures for differentially private streams and iteratively pruned deep neural networks.

DNS cache timing attacks for continuous user behavior monitoring, SMTP timeout manipulation for amplification attacks, co-training for dynamic password guessing, and performance-counter side channels on AMD SEV-SNP VMs.

Infostealer data-driven analysis, LLM honeypots, persistent billable state DoW, bypassing unlearning via tokenization, DistillGuard for NPM, MoSign VR authentication, AgentKernel native trust, and Byzantine-robust intrusion detection.

Today’s preprint highlights include unprivileged IMU side-channels on Apple Silicon, TEE-backed digital rights architectures, hardware-rooted Linux anti-cheat, provenance-aware transformers against indirect prompt injection, and cryptographic implementation benchmarking for LLMs.

A systematic evaluation of cache side-channel attacks on modern Intel microarchitectures, and a new security architecture to mitigate integrity violations in LLM-integrated app systems.