Text‑Vector Integrity Gap Enables Hijacking in Offline Retrieval‑Augmented Generation

# Text‑Vector Integrity Gap Enables Hijacking in Offline Retrieval‑Augmented Generation

https://gcdradar.vercel.app/incidents/text-vector-integrity-gap-enables-hijacking-in-offline-retri · AIR-2026-1788 · live

- Lane: Research
- Severity: 1 (Research / proof of concept): demonstrates a novel attack on RAG infrastructure; no known exploitation
- Fix status: Fix status unknown
- First reported: 2026-11-05
- AI role: Attack on AI

## Summary

The paper identifies a text‑vector integrity gap in offline retrieval‑augmented generation where signed knowledge packs can be manipulated to control query results. A verifier that tolerates vector deviations enables an attacker to hijack a share of queries.

## What to do

No action recorded; read the sources below and the vendor's advisory.

## Latest updates

- 2026-11-05: Signed but Steerable: Measuring and Closing the Text-Vector Integrity Gap in Offline Retrieval-Augmented Generation (https://arxiv.org/abs/2611.06498)

## Sources

- 2026-11-05 arXiv: Signed but Steerable: Measuring and Closing the Text-Vector Integrity Gap in Offline Retrieval-Augmented Generation https://arxiv.org/abs/2611.06498

Verify against the vendor's advisory before changing production. Data CC BY-SA 4.0, https://gcdradar.vercel.app.