Cybersecurity Data Extraction from Common Crawl

This collection of recent papers highlights multiple practical and research-scale threats and defenses across ML and cyber systems. Key contributions include: Alpha-Root, a Common Crawl–derived cybersecurity dataset; HubScan, an open-source scanner for hubness poisoning in RAG/vector-search systems (FAISS, Pinecone, Qdrant, Weaviate); empirical studies showing prompt-injection and jailbreak variability across open LLMs and that lightweight inference-time filters are routinely bypassed by long, reasoning-heavy prompts; stealthy training-data poisoning and label-flip attacks (Poisoned Acoustics)