{"paper":{"title":"HalluScan: A Systematic Benchmark for Detecting and Mitigating Hallucinations in Instruction-Following LLMs","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Systematic benchmarking reveals NLI Verification as the most effective method for detecting hallucinations in LLMs at an AUROC of 0.88.","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Ahmed Cherif","submitted_at":"2026-05-04T10:43:27Z","abstract_excerpt":"Large Language Models (LLMs) have demonstrated remarkable capabilities across diverse natural language processing tasks, yet they remain susceptible to hallucinations -- generating content that is factually incorrect, unfaithful to provided context, or misaligned with user instructions. We present HalluScan, a comprehensive benchmark framework that systematically evaluates hallucination detection and mitigation across 72 configurations spanning 6 detection methods, 4 open-weight model families, and 3 diverse domains. We introduce three key contributions: (1) HalluScore, a novel composite metri"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"NLI Verification achieves the highest overall AUROC of 0.88, while RAV achieves the second-highest AUROC of 0.66. HalluScore achieves a Pearson correlation of r = 0.41 with human expert judgments.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the 72 configurations, chosen models, and three domains sufficiently represent the full range of hallucination behaviors in instruction-following LLMs and that human judgments provide a stable ground truth for validating the new metric.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"HalluScan benchmark tests hallucination detectors on LLMs, identifies NLI Verification as top performer with 0.88 AUROC, and introduces HalluScore (r=0.41 with humans) plus a routing method for 2x cost savings.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Systematic benchmarking reveals NLI Verification as the most effective method for detecting hallucinations in LLMs at an AUROC of 0.88.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"f3a18cc1848e370773a4dc20497832d0bf7bf8819d4796e7cdab01b6c775c0d7"},"source":{"id":"2605.02443","kind":"arxiv","version":2},"verdict":{"id":"b7d1684b-537a-4e18-bf2b-4b834e1c50ee","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-08T18:25:03.167234Z","strongest_claim":"NLI Verification achieves the highest overall AUROC of 0.88, while RAV achieves the second-highest AUROC of 0.66. HalluScore achieves a Pearson correlation of r = 0.41 with human expert judgments.","one_line_summary":"HalluScan benchmark tests hallucination detectors on LLMs, identifies NLI Verification as top performer with 0.88 AUROC, and introduces HalluScore (r=0.41 with humans) plus a routing method for 2x cost savings.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the 72 configurations, chosen models, and three domains sufficiently represent the full range of hallucination behaviors in instruction-following LLMs and that human judgments provide a stable ground truth for validating the new metric.","pith_extraction_headline":"Systematic benchmarking reveals NLI Verification as the most effective method for detecting hallucinations in LLMs at an AUROC of 0.88."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.02443/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"ai_meta_artifact","ran_at":"2026-05-20T15:40:33.758746Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-20T03:31:22.113923Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T16:22:47.877602Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"b8c5b882ae4b8996fd4da6824b4d1c15004b86d4411de55f572f5a396c4a5592"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":3,"snapshot_sha256":"439e5e5d84d3042ab501687000b3249d1ced253ce7681d07e180d3c7a4a204bb"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}