Pith. sign in

REVIEW 12 cited by

Poisoning Retrieval Corpora by Injecting Adversarial Passages

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2310.19156 v1 pith:HQJAD6FE submitted 2023-10-29 cs.CL cs.IR

Poisoning Retrieval Corpora by Injecting Adversarial Passages

classification cs.CL cs.IR
keywords passagesretrievaladversarialattackdensequeriessystemscorpora
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
read the original abstract

Dense retrievers have achieved state-of-the-art performance in various information retrieval tasks, but to what extent can they be safely deployed in real-world applications? In this work, we propose a novel attack for dense retrieval systems in which a malicious user generates a small number of adversarial passages by perturbing discrete tokens to maximize similarity with a provided set of training queries. When these adversarial passages are inserted into a large retrieval corpus, we show that this attack is highly effective in fooling these systems to retrieve them for queries that were not seen by the attacker. More surprisingly, these adversarial passages can directly generalize to out-of-domain queries and corpora with a high success attack rate -- for instance, we find that 50 generated passages optimized on Natural Questions can mislead >94% of questions posed in financial documents or online forums. We also benchmark and compare a range of state-of-the-art dense retrievers, both unsupervised and supervised. Although different systems exhibit varying levels of vulnerability, we show they can all be successfully attacked by injecting up to 500 passages, a small fraction compared to a retrieval corpus of millions of passages.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 12 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Your Agent's Memories Are Not Its Own: Forged Reasoning Attacks on LLM Agent Memory and Defenses

    cs.CR 2026-07 conditional novelty 7.0

    FARMA forges and self-amplifies an agent's reasoning history with evasive language to induce unsafe skips; SENTINEL's Reasoning Guard reduces ASR to 0% across tested agents and models.

  2. Replicating Belief, Not Bits: Epistemic State Replication for Agentic Systems

    cs.AI 2026-07 conditional novelty 7.0

    ESR separates an immutable evidence log from a stochastic belief lineage so agent replicas stay semantically compatible without bitwise state equality.

  3. Embedding Inference Attack

    cs.CR 2026-07 unverdicted novelty 7.0

    Tailored queries enable identification of the embedding model used by a black-box IR system from the unordered set of retrieved documents, even when a reranker is present.

  4. Needle-in-RAG: Prompt-Conditioned Character-Level Traceback of Poisoned Spans in Retrieved Evidence

    cs.CR 2026-05 unverdicted novelty 7.0

    RAGCharacter localizes poisoned character spans in RAG evidence via prompt-conditioned counterfactual masking and achieves the best accuracy-over-attribution trade-off across tested attacks and models.

  5. Led to Mislead: Adversarial Content Injection for Attacks on Neural Ranking Models

    cs.IR 2026-05 unverdicted novelty 7.0

    CRAFT is a supervised LLM framework using retrieval-augmented generation, self-refinement, fine-tuning, and preference optimization to create fluent adversarial content that boosts target ranks in neural ranking model...

  6. Agent Security Needs Redefinition through a Holistic Framework

    cs.CR 2026-07 conditional novelty 6.0

    Agent security should be redefined around four contextual authorization properties instead of the content of the action performed.

  7. Selection Integrity for LLM Graph Memory: An Accumulability Criterion for Information-Flow-Blind Retrieval

    cs.CR 2026-06 unverdicted novelty 6.0

    Provenance checks in graph memory are blind to structural attacks that reallocate top-k membership; authselect prevents this by enforcing selection on the authenticated subgraph only.

  8. Detecting Is Not Resolving: The Monitoring Control Gap in Retrieval Augmented LLMs

    cs.AI 2026-05 unverdicted novelty 6.0

    RAG models exhibit a monitoring-control gap: they acknowledge epistemic conflicts in accumulating documents yet fail to constrain unsafe recommendations, with single-turn tests overestimating safety.

  9. AttnTrace: Contextual Attribution of Prompt Injection and Knowledge Corruption

    cs.CL 2025-08 unverdicted novelty 6.0

    AttnTrace is an attention-weight-based context traceback method for LLMs that claims higher accuracy and efficiency than prior art like TracLLM while aiding prompt injection detection.

  10. ToE: A Hierarchical and Explainable Claim Verification Framework with Dynamic Multi-source Evidence Retrieval and Aggregation

    cs.AI 2026-06 unverdicted novelty 5.0

    ToE is a hierarchical claim verification framework using RL-driven multi-source retrieval, evidence evaluation, and tree aggregation that reports 4-24 point gains over baselines especially on poisoned inputs.

  11. Through the Stealth Lens: Attention-Aware Defenses Against Poisoning in RAG

    cs.CR 2025-06 unverdicted novelty 5.0

    Introduces NPAS and AV Filter using LLM attention weights to defend RAG against poisoning, reporting up to 20% accuracy gains while adaptive attacks reach 35% success.

  12. Agentic AI Security: Threats, Defenses, Evaluation, and Open Challenges

    cs.AI 2025-10 unverdicted novelty 4.0

    A survey that taxonomizes threats to agentic AI, reviews benchmarks and evaluation methods, discusses technical and governance defenses, and identifies open challenges.