Research map / Adversarial robustness and security / Backdoor and poisoning
Fact2Fiction: Targeted Poisoning Attack to Agentic Fact-checking System
Haorui He, Yupeng Li, Bin Benjamin Zhu, Dacheng Wen, Reynold Cheng, Francis C. M. Lau
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use Fact2Fiction instead of PoisonedRAG | attack on DEFAME, 1% poison · AVeriTeC dev | ASR (%): 33.5 -> 42.4 | Improved |
| use Fact2Fiction instead of PoisonedRAG | attack on Simple RAG, 1% poison · AVeriTeC dev | ASR (%): 42.4 -> 43.4 | No clear effect |
| add answer planning using justifications (ablation) | Fact2Fiction on DEFAME, 1% · AVeriTeC dev | ASR (%): 40.9 -> 42.4 | Improved |
| add budget planning (ablation) | Fact2Fiction on DEFAME, 1% · AVeriTeC dev | ASR (%): 34.6 -> 42.4 | Improved |
| add budget planning (ablation) | Fact2Fiction on InFact, 8% · AVeriTeC dev | ASR (%): 56.9 -> 59.9 | Improved |
| add query planning (ablation) | Fact2Fiction on DEFAME, 8% · AVeriTeC dev | ASR (%): 49.4 -> 63.6 | Improved |
| add query planning (ablation) | Fact2Fiction on Simple RAG, 8% · AVeriTeC dev | ASR (%): 61.5 -> 65.7 | Improved |
| target agentic system instead of simple RAG | PoisonedRAG, 8% · AVeriTeC dev | ASR (%): 57.4 -> 42.4 (DEFAME), 45.3 (InFact) | Hurt |
| apply paraphrasing defense | Fact2Fiction on DEFAME, 1% · AVeriTeC dev | ASR (%): 42.4 -> 39.4 | Hurt |
| apply malicious detection defense (k-means) | PoisonedRAG vs Fact2Fiction, DEFAME 1% · AVeriTeC dev | ASR (%): PoisonedRAG 33.5 -> 10.0; Fact2Fiction 42.4 -> 26.0 | Hurt |
| increase poison rate for Naive/Prompt Injection | Naive attack on DEFAME · AVeriTeC dev | ASR (%): 17.8 -> 18.2 (1% -> 8%) | No clear effect |
Closest papers by meaning
- DECEIVE-AFC: Adversarial Claim Attacks against Search-Enabled LLM-based Fact-Checking Systems (NeurIPS 2026 · similarity 0.85)
- Goal-Aware Identification and Rectification of Misinformation in Multi-Agent Systems (ICLR 2026 · similarity 0.85)
- BiDeV: Bilateral Defusing Verification for Complex Claim Fact-Checking (AAAI 2025 · similarity 0.82)
- Beyond Detection: Exploring Evidence-based Multi-Agent Debate for Misinformation Intervention and Persuasion (AAAI 2026 · similarity 0.81)
- AgentHarm: A Benchmark for Measuring Harmfulness of LLM Agents (ICLR 2025 · similarity 0.81)
- Monitoring Decomposition Attacks with Lightweight Sequential Monitors (ICLR 2026 · similarity 0.81)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
