MIRAGE: Defending Long-Form RAG Against Misinformation Pollution
For practitioners of long-form RAG, MIRAGE provides a training-free defense against realistic retrieval pollution, a critical problem for deploying LLMs in knowledge-intensive tasks.
MIRAGE defends long-form RAG against misinformation pollution by using an NLI-based claim graph and a Defended-Claims Gate, restoring factuality under polluted evidence and outperforming prior robust-RAG methods across four benchmarks.
Retrieval-Augmented Generation (RAG) improves factuality by grounding LLMs in external evidence, but real-world retrieval is often polluted: semantically relevant passages may contain subtle misinformation, misleading framings, or fabrications. We introduce MIRAGE, a training-free, model-agnostic defense for long-form RAG. MIRAGE builds an NLI-based cross-document claim graph and applies a Defended-Claims Gate to either condition generation on a consistent, multi-source supported subset or to block retrieval and answer parametrically. We also release a minimal-edit pollution protocol spanning four perturbation families (Unambiguous, Conflicting, Misleading, Fabricated) to construct matched clean, mixed, and fully polluted evaluation regimes. Across four long-form QA benchmarks and multiple commercial and open-weight LLMs, pollution severely degrades vanilla RAG, while MIRAGE consistently restores factuality under mixed and fully polluted evidence and outperforms prior robust-RAG methods. Our implementation and datasets are available at https://github.com/SaadElDine/MIRAGE.