RAG Security - Vector & Embedding Weaknesses
Details
Event #1 — Vector & Embedding Weaknesses (RAG security)
Kicking off the Bay Area AI Security Research Meetup with a deep-dive discussion on LLM08 — Vector and Embedding Weaknesses, one of the most under-discussed entries in the OWASP Top 10 for LLM Applications.
If you're building anything with RAG, vector databases, or embedding pipelines, this one's for you. We'll dig into the real attack surface: embedding inversion, cross-tenant / namespace leakage in shared vector stores, data poisoning of long-term memory, and the federation knowledge conflicts that quietly break retrieval trust. Less theory, more "here's how it actually breaks."
This is a discussion, not a lecture. Come with a question you're stuck on, a finding you want a second pair of eyes on, or a defense you think holds up — and we'll pressure-test it together. Small group, so everyone in the room gets airtime.
No prep required, but if you want to skim beforehand:
- OWASP LLM08 entry: https://genai.owasp.org/llmrisk/llm082025-vector-and-embedding-weaknesses/
Security engineers, ML folks, and researchers all welcome. Want to co-lead part of the discussion or suggest future topics? Email [[aisecurityinbay@proton.me](mailto:aisecurityinbay@proton.me)]