The nightmare of securing RAG pipelines in production... thoughts?
retained text
Honestly it feels like every dev team is rushing to spin up internal LLMs and RAG pipelines, but the security side is just a complete wild west right now. Traditional WAFs and DLP solutions definetly aren't cutting it when it comes to prompt injection or potential data leaks thru vector DBs. How are you guys actually handling security for this? Relying on semantic firewalls, building custom middleware, or just locking down the context window and hoping for the best? Would love to hear what's realistically working for you guys because I'm loosing my mind a bit trying to lock this down.
document fetched 2026-10-15 00:25 · canonical https://reddit.com/r/cybersecurity/comments/1x5u2qc/the_nightmare_of_securing_rag_pipelines_in