Emerging Evaluations ProjectEmpirical security research on AI model vulnerabilities.https://emergingevaluations.com/en-usBlind Sentinel: RAG Poisoning and the Limits of Retrieval as a Security Controlhttps://emergingevaluations.com/reports/blind-sentinal/https://emergingevaluations.com/reports/blind-sentinal/An empirical study of indirect prompt injection via RAG poisoning in a SOC analyst scenario, finding that retrieval — not model alignment — is the load-bearing security control.Wed, 15 Jan 2025 00:00:00 GMTLLM01LLM06