{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:DHA4TDJ7NO65Z2KWWF6Z7SSNUY","short_pith_number":"pith:DHA4TDJ7","schema_version":"1.0","canonical_sha256":"19c1c98d3f6bbddce956b17d9fca4da6201aca5f9516fa6bef9f2ebe4ae88713","source":{"kind":"arxiv","id":"2306.04528","version":5},"attestation_state":"computed","paper":{"title":"PromptRobust: Towards Evaluating the Robustness of Large Language Models on Adversarial Prompts","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CR","cs.LG"],"primary_cat":"cs.CL","authors_text":"Hao Chen, Jiaheng Zhou, Jindong Wang, Kaijie Zhu, Linyi Yang, Neil Zhenqiang Gong, Wei Ye, Xing Xie, Yidong Wang, Yue Zhang, Zichen Wang","submitted_at":"2023-06-07T15:37:00Z","abstract_excerpt":"The increasing reliance on Large Language Models (LLMs) across academia and industry necessitates a comprehensive understanding of their robustness to prompts. In response to this vital need, we introduce PromptRobust, a robustness benchmark designed to measure LLMs' resilience to adversarial prompts. This study uses a plethora of adversarial textual attacks targeting prompts across multiple levels: character, word, sentence, and semantic. The adversarial prompts, crafted to mimic plausible user errors like typos or synonyms, aim to evaluate how slight deviations can affect LLM outcomes while "},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2306.04528","kind":"arxiv","version":5},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2023-06-07T15:37:00Z","cross_cats_sorted":["cs.CR","cs.LG"],"title_canon_sha256":"e822ca7997975f697e764af09fa5684f2331a852d213b99bb9f6d785f7f14fdc","abstract_canon_sha256":"cd68adc0f293a6c59ff16d2b81226e28d7f2b9f8b767ca84089d06c2e985c0e4"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:44:09.080869Z","signature_b64":"T3swW72iFMg7ZcoJGwnHs/iomprZAwgqUHcrcpiIrz2E5p8XWYdiDIDsT4PRUenBcUpQZMZeRP1stFVEgCsCDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"19c1c98d3f6bbddce956b17d9fca4da6201aca5f9516fa6bef9f2ebe4ae88713","last_reissued_at":"2026-07-05T08:44:09.080457Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:44:09.080457Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"PromptRobust: Towards Evaluating the Robustness of Large Language Models on Adversarial Prompts","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CR","cs.LG"],"primary_cat":"cs.CL","authors_text":"Hao Chen, Jiaheng Zhou, Jindong Wang, Kaijie Zhu, Linyi Yang, Neil Zhenqiang Gong, Wei Ye, Xing Xie, Yidong Wang, Yue Zhang, Zichen Wang","submitted_at":"2023-06-07T15:37:00Z","abstract_excerpt":"The increasing reliance on Large Language Models (LLMs) across academia and industry necessitates a comprehensive understanding of their robustness to prompts. In response to this vital need, we introduce PromptRobust, a robustness benchmark designed to measure LLMs' resilience to adversarial prompts. This study uses a plethora of adversarial textual attacks targeting prompts across multiple levels: character, word, sentence, and semantic. The adversarial prompts, crafted to mimic plausible user errors like typos or synonyms, aim to evaluate how slight deviations can affect LLM outcomes while "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2306.04528","kind":"arxiv","version":5},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2306.04528/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2306.04528","created_at":"2026-07-05T08:44:09.080512+00:00"},{"alias_kind":"arxiv_version","alias_value":"2306.04528v5","created_at":"2026-07-05T08:44:09.080512+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2306.04528","created_at":"2026-07-05T08:44:09.080512+00:00"},{"alias_kind":"pith_short_12","alias_value":"DHA4TDJ7NO65","created_at":"2026-07-05T08:44:09.080512+00:00"},{"alias_kind":"pith_short_16","alias_value":"DHA4TDJ7NO65Z2KW","created_at":"2026-07-05T08:44:09.080512+00:00"},{"alias_kind":"pith_short_8","alias_value":"DHA4TDJ7","created_at":"2026-07-05T08:44:09.080512+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":24,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.23716","citing_title":"Legal Reasoning Is Not Lawyering: Rethinking Legal Benchmarks for Pro Se Access to Justice","ref_index":17,"is_internal_anchor":false},{"citing_arxiv_id":"2606.12978","citing_title":"Trajectory-Level Redirection Attacks on Vision-Language-Action Models","ref_index":43,"is_internal_anchor":false},{"citing_arxiv_id":"2606.06924","citing_title":"From Sampled Outcomes to Capability Distributions: Rethinking Supervision for LLM Routing","ref_index":158,"is_internal_anchor":false},{"citing_arxiv_id":"2606.01210","citing_title":"Can we trust LLM Self-Explanations for Entity Resolution?","ref_index":73,"is_internal_anchor":false},{"citing_arxiv_id":"2606.01441","citing_title":"Dive into Ambiguity: A*-Inspired Multi-Agents Commonsense Obfuscation Attack on LLM Prompts","ref_index":43,"is_internal_anchor":false},{"citing_arxiv_id":"2307.15043","citing_title":"Universal and Transferable Adversarial Attacks on Aligned Language Models","ref_index":29,"is_internal_anchor":false},{"citing_arxiv_id":"2402.06922","citing_title":"Whispers in the Machine: Confidentiality in Agentic Systems","ref_index":30,"is_internal_anchor":false},{"citing_arxiv_id":"2409.10102","citing_title":"Trustworthiness in Retrieval-Augmented Generation Systems: A Survey","ref_index":92,"is_internal_anchor":false},{"citing_arxiv_id":"2406.04244","citing_title":"Benchmark Data Contamination of Large Language Models: A Survey","ref_index":191,"is_internal_anchor":false},{"citing_arxiv_id":"2604.23135","citing_title":"Characterizing Paraphrase-Induced Failures in Lean 4 Autoformalization","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2401.05561","citing_title":"TrustLLM: Trustworthiness in Large Language Models","ref_index":172,"is_internal_anchor":false},{"citing_arxiv_id":"2309.08532","citing_title":"EvoPrompt: Connecting LLMs with Evolutionary Algorithms Yields Powerful Prompt Optimizers","ref_index":137,"is_internal_anchor":false},{"citing_arxiv_id":"2603.03332","citing_title":"Fragile Thoughts: How Large Language Models Handle Chain-of-Thought Perturbations","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2603.10477","citing_title":"PEEM: Prompt Engineering Evaluation Metrics for Interpretable Joint Evaluation of Prompts and Responses","ref_index":43,"is_internal_anchor":false},{"citing_arxiv_id":"2309.00614","citing_title":"Baseline Defenses for Adversarial Attacks Against Aligned Language Models","ref_index":65,"is_internal_anchor":false},{"citing_arxiv_id":"2604.16421","citing_title":"Measuring Representation Robustness in Large Language Models for Geometry","ref_index":40,"is_internal_anchor":false},{"citing_arxiv_id":"2309.01219","citing_title":"Siren's Song in the AI Ocean: A Survey on Hallucination in Large Language Models","ref_index":30,"is_internal_anchor":false},{"citing_arxiv_id":"2605.09041","citing_title":"BiAxisAudit: A Novel Framework to Evaluate LLM Bias Across Prompt Sensitivity and Response-Layer Divergence","ref_index":34,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10516","citing_title":"Consistency as a Testable Property: Statistical Methods to Evaluate AI Agent Reliability","ref_index":35,"is_internal_anchor":false},{"citing_arxiv_id":"2605.04665","citing_title":"Paraphrase-Induced Output-Mode Collapse: When LLMs Break Character Under Semantically Equivalent Inputs","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2604.24712","citing_title":"When Prompt Under-Specification Improves Code Correctness: An Exploratory Study of Prompt Wording and Structure Effects on LLM-Based Code Generation","ref_index":46,"is_internal_anchor":false},{"citing_arxiv_id":"2604.23338","citing_title":"A Systematic Survey of Security Threats and Defenses in LLM-Based AI Agents: A Layered Attack Surface Framework","ref_index":77,"is_internal_anchor":false},{"citing_arxiv_id":"2604.23135","citing_title":"Characterizing Paraphrase-Induced Failures in Lean 4 Autoformalization","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2605.04665","citing_title":"Paraphrase-Induced Output-Mode Collapse: When LLMs Break Character Under Semantically Equivalent Inputs","ref_index":5,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/DHA4TDJ7NO65Z2KWWF6Z7SSNUY","json":"https://pith.science/pith/DHA4TDJ7NO65Z2KWWF6Z7SSNUY.json","graph_json":"https://pith.science/api/pith-number/DHA4TDJ7NO65Z2KWWF6Z7SSNUY/graph.json","events_json":"https://pith.science/api/pith-number/DHA4TDJ7NO65Z2KWWF6Z7SSNUY/events.json","paper":"https://pith.science/paper/DHA4TDJ7"},"agent_actions":{"view_html":"https://pith.science/pith/DHA4TDJ7NO65Z2KWWF6Z7SSNUY","download_json":"https://pith.science/pith/DHA4TDJ7NO65Z2KWWF6Z7SSNUY.json","view_paper":"https://pith.science/paper/DHA4TDJ7","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2306.04528&json=true","fetch_graph":"https://pith.science/api/pith-number/DHA4TDJ7NO65Z2KWWF6Z7SSNUY/graph.json","fetch_events":"https://pith.science/api/pith-number/DHA4TDJ7NO65Z2KWWF6Z7SSNUY/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/DHA4TDJ7NO65Z2KWWF6Z7SSNUY/action/timestamp_anchor","attest_storage":"https://pith.science/pith/DHA4TDJ7NO65Z2KWWF6Z7SSNUY/action/storage_attestation","attest_author":"https://pith.science/pith/DHA4TDJ7NO65Z2KWWF6Z7SSNUY/action/author_attestation","sign_citation":"https://pith.science/pith/DHA4TDJ7NO65Z2KWWF6Z7SSNUY/action/citation_signature","submit_replication":"https://pith.science/pith/DHA4TDJ7NO65Z2KWWF6Z7SSNUY/action/replication_record"}},"created_at":"2026-07-05T08:44:09.080512+00:00","updated_at":"2026-07-05T08:44:09.080512+00:00"}