Extracting website fingerprinting traces across positions, exploiting time synchronization in 802.11 for covert channels, breaking LLM watermarks with character perturbations, and characterizing censorship implementation in Chinese LLMs.
New perspectives on LLM cryptographic capabilities, agent oversight bypasses, and hardware fuzzing paradigms.
New vulnerabilities in eBPF, WebGPU fuzzing, and backdoor attacks on LLM-powered robots.
New findings spanning LLM cache poisoning, npm prototype pollution, and kernel compartmentalization with PKS.
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.