Dimensions overall score 9.0

GitHub Code Pulse

No public code linked for this paper yet.

• HubScan is an open-source security scanner designed to identify hubs in RAG systems by evaluating vector indices and embeddings.technical 95%
• HubScan employs a multi-detector architecture that includes statistical hubness detection, cluster spread analysis, stability testing under query perturbations, and domain/modality-aware detection.method 95%
• HubScan supports multiple vector databases, including FAISS, Pinecone, Qdrant, and Weaviate, and offers versatile retrieval techniques like vector similarity, hybrid search, and lexical matching with reranking.technical 95%
• HubScan achieves 90% recall at a 0.2% alert budget and 100% recall at 0.4% on adversarial hubness benchmarks.result 90%
• Domain-scoped scanning with HubScan recovers 100% of targeted attacks that evade global detection.result 90%
• Production validation on 1M real web documents from MS MARCO demonstrates significant score separation between clean documents and adversarial content using HubScan.result 90%
• HubScan might face challenges with evolving adversarial tactics or new attack forms that bypass current detection methods, requiring continuous updates.limitation 80%
• There is a market need to secure RAG systems in enterprises using AI for decision support, which creates a large market opportunity in the cybersecurity sector.market 70%
• HubScan utilizes a multi-detector architecture for identifying hubs in RAG systems, integrating statistical hubness detection, cluster spread analysis, stability testing, and domain/modality-aware detection.method 70%
• HubScan achieves 90% recall at a 0.2% alert budget and 100% recall at 0.4% on adversarial hubness benchmarks.result 70%
• HubScan supports integration with multiple vector databases including FAISS, Pinecone, Qdrant, and Weaviate.technical 70%
• Domain-scoped scanning by HubScan recovers 100% of targeted attacks that evade global detection.result 70%

Competitive landscape

Competitor map is still being generated for this paper. Enable generation or check back soon.

Startup potential card