Arxiv benchmark for polymorphic Sybil poisoning in RAG, with forced exposure and a 4-way failure taxonomy instead of yet another toy attack.

Native Goldilocks/G64 NTTs on an RTX4090, with qingming_fast/standard benchmarks and validated LDE domains up to 2^30.

ToolPrivacyBench, a benchmark for purpose-bound privacy in tool-using LLM agents, because apparently prompts needed audits.