Influence Factors on RAG Poisoning
2026-06-12T07:23:31Z•36433ec73903d67e5fa8a1cee8076a0d51c15d0f5cd95cd387986f4f2281dacf
Amnesia attackCAPEDHMAC provenanceLFPMRAG poisoningSMSRauditable telemetrybackdoor mitigationcontinual learningfeature-space defensehard-negative augmentationmemory poisoningmobile GUI privacymodel mergingmulti-session memory poisoningneighbor leakage ratepoisoning attacksrandomised ablation voting','prompt injection','PI-Hunter','red‑replay attacksretrieval depthretrieval-augmented generationretriever architectureselective screenshot exposuretrigger leakagevision-language agents
What happened
Collection of recent arXiv papers (2026-06-12) surveying new attack surfaces and defenses in modern ML/agent systems. Key contributions include: an extensive factorial study of poisoning exposure in Retrieval-Augmented Generation (RAG) systems showing retriever architecture and retrieval depth drive vulnerability; LFPM, a feature-space backdoor mitigation method for model-merging that preserves clean-task performance; formalization and measurement of "trigger leakage" in vision-language agentic systems and mitigation via hard-negative sampling; Amnesia, an index-only replay-composition attack
Why it matters
A reviewed impact interpretation has not been published for this record.
Evidence and limitations
- Source ID
- arxiv_cs_cr
- Record identifier
- 36433ec73903d67e5fa8a1cee8076a0d51c15d0f5cd95cd387986f4f2281dacf
- Enrichment time
- 2026-06-12T07:23:31Z
- AI-assisted enrichment
- Yes
This record may overlap with other records. Its enrichment can be incomplete or wrong, and machine assistance was used. Validate consequential decisions against the linked source and your own environment.