threatdeck
Live 14:45 UTC ·

New Technique Medium TD-2026-0081 For AI Eng, AppSec, CISO AI-assisted

SNAPRAG: Adversarial Query Attack Extracts RAG Knowledge Bases at Lower Query Cost

The paper presents SNAPRAG, a black-box attack that extracts private RAG knowledge bases using algorithmically generated adversarial queries without external LLMs. The authors report it needs up to 6.77 times fewer queries than existing approaches to extract half of a target knowledge base. Within a fixed query budget, it recovers up to 96.9% of the knowledge base.

Why it matters

Organizations running RAG systems over private documents may have their proprietary knowledge bases extracted through repeated querying alone. Lower query cost makes such extraction cheaper and harder to detect by rate limits.

Do this

  1. Rate-limit and monitor query volume and similarity patterns against RAG endpoints to detect systematic knowledge base probing.
  2. Restrict exposed RAG responses to limit retrieved text and metadata returned to end users.
  3. Evaluate RAG deployments against extraction benchmarks, including the defended targets described in the paper, before exposing sensitive corpora.
  4. Classify and minimize the sensitivity of documents indexed in retrieval stores where full extraction would be harmful.

Signals

  • NoCISA KEV
  • 23/100Priority
  • 60%Confidence

Evidence

“SNAPRAG uses a discrete token-mutation algorithm to generate novel queries in a strict black-box setting.”

arXiv (AI security papers)

“to extract half of the target knowledge base, our method requires up to 6.77$\times$ fewer queries than existing approaches.”

arXiv (AI security papers)

Sources

Affected / related

SNAPRAG, RAG knowledge base extraction

Published 2026-11-09 11:30 UTC · updated 2026-11-12 18:05 UTC · AI-assisted summary, checked against quoted sources · How we score

Following

Follow the vendors, products, AI tools, techniques and actors you care about. Matching stories, CVEs and exploited bugs are marked, a Following tab appears in the feed, and returning visits show what changed. Saved only in this browser.

Presets

Get updates

A personal RSS feed of new stories on everything you follow. Paste it into any feed reader, Slack or Teams RSS app.