{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:YDOTUXDKCOJ2VSOKV7ENL4O5QY","short_pith_number":"pith:YDOTUXDK","schema_version":"1.0","canonical_sha256":"c0dd3a5c6a1393aac9caafc8d5f1dd8627b36a6125e5d58357767e8562fdbf91","source":{"kind":"arxiv","id":"2311.03348","version":2},"attestation_state":"computed","paper":{"title":"Scalable and Transferable Black-Box Jailbreaks for Language Models via Persona Modulation","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Arush Tagade, Javier Rando, Quentin Feuillade--Montixi, Rusheb Shah, Soroush Pour, Stephen Casper","submitted_at":"2023-11-06T18:55:18Z","abstract_excerpt":"Despite efforts to align large language models to produce harmless responses, they are still vulnerable to jailbreak prompts that elicit unrestricted behaviour. In this work, we investigate persona modulation as a black-box jailbreaking method to steer a target model to take on personalities that are willing to comply with harmful instructions. Rather than manually crafting prompts for each persona, we automate the generation of jailbreaks using a language model assistant. We demonstrate a range of harmful completions made possible by persona modulation, including detailed instructions for syn"},"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":"2311.03348","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.CL","submitted_at":"2023-11-06T18:55:18Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"7a0857d817bb2bb942d775ca81034a98fce6901ada1df14bf425bfbc9de18366","abstract_canon_sha256":"b8ccf03db52eb7d8aa86d7fd7a0115626fd11618248eed8d1bfcc2f8dfb49500"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:16:02.521924Z","signature_b64":"/NNZfPRQ/l/f98b3IGAWAD2frNbN2DyPoExb/hq9PFz/7ZHyxScOnwknXhbfRrhCjXnDwdcez6YZnxex+sATBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c0dd3a5c6a1393aac9caafc8d5f1dd8627b36a6125e5d58357767e8562fdbf91","last_reissued_at":"2026-07-05T07:16:02.521459Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:16:02.521459Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Scalable and Transferable Black-Box Jailbreaks for Language Models via Persona Modulation","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Arush Tagade, Javier Rando, Quentin Feuillade--Montixi, Rusheb Shah, Soroush Pour, Stephen Casper","submitted_at":"2023-11-06T18:55:18Z","abstract_excerpt":"Despite efforts to align large language models to produce harmless responses, they are still vulnerable to jailbreak prompts that elicit unrestricted behaviour. In this work, we investigate persona modulation as a black-box jailbreaking method to steer a target model to take on personalities that are willing to comply with harmful instructions. Rather than manually crafting prompts for each persona, we automate the generation of jailbreaks using a language model assistant. We demonstrate a range of harmful completions made possible by persona modulation, including detailed instructions for syn"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2311.03348","kind":"arxiv","version":2},"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/2311.03348/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":"2311.03348","created_at":"2026-07-05T07:16:02.521515+00:00"},{"alias_kind":"arxiv_version","alias_value":"2311.03348v2","created_at":"2026-07-05T07:16:02.521515+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2311.03348","created_at":"2026-07-05T07:16:02.521515+00:00"},{"alias_kind":"pith_short_12","alias_value":"YDOTUXDKCOJ2","created_at":"2026-07-05T07:16:02.521515+00:00"},{"alias_kind":"pith_short_16","alias_value":"YDOTUXDKCOJ2VSOK","created_at":"2026-07-05T07:16:02.521515+00:00"},{"alias_kind":"pith_short_8","alias_value":"YDOTUXDK","created_at":"2026-07-05T07:16:02.521515+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":20,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.25476","citing_title":"A Red Teaming Framework for Large Language Models: A Case Study on Faithfulness Evaluation","ref_index":35,"is_internal_anchor":false},{"citing_arxiv_id":"2607.01277","citing_title":"Cognitive Firewall: A Proactive, Zero-Trust, Multi-Gate Framework for LLM Safety","ref_index":38,"is_internal_anchor":false},{"citing_arxiv_id":"2606.05609","citing_title":"SlotGCG: Exploiting the Positional Vulnerability in LLMs for Jailbreak Attacks","ref_index":52,"is_internal_anchor":false},{"citing_arxiv_id":"2606.05523","citing_title":"CHASE: Adversarial Red-Blue Teaming for Improving LLM Safety using Reinforcement Learning","ref_index":48,"is_internal_anchor":false},{"citing_arxiv_id":"2605.24279","citing_title":"ContextEcho: A Benchmark for Persona Drift in Long Agentic-Coding Sessions","ref_index":68,"is_internal_anchor":false},{"citing_arxiv_id":"2606.00801","citing_title":"Quality-Diversity Evolution for Discovering Diverse Vulnerabilities in LLM Safety","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2312.03853","citing_title":"Dr. Jekyll and Mr. Hyde: Two Faces of LLMs","ref_index":17,"is_internal_anchor":false},{"citing_arxiv_id":"2508.20325","citing_title":"GUARD: Guideline Upholding Test through Adaptive Role-play and Jailbreak Diagnostics for LLMs","ref_index":29,"is_internal_anchor":false},{"citing_arxiv_id":"2509.06701","citing_title":"Probabilistic Modeling of Latent Agentic Substructures in Deep Neural Networks","ref_index":73,"is_internal_anchor":false},{"citing_arxiv_id":"2402.10260","citing_title":"A StrongREJECT for Empty Jailbreaks","ref_index":29,"is_internal_anchor":false},{"citing_arxiv_id":"2601.03416","citing_title":"GAMBIT: A Gamified Jailbreak Framework for Multimodal Large Language Models","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2602.11528","citing_title":"Stop Tracking Me! Proactive Defense Against Attribute Inference Attack in LLMs","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2404.01318","citing_title":"JailbreakBench: An Open Robustness Benchmark for Jailbreaking Large Language Models","ref_index":44,"is_internal_anchor":false},{"citing_arxiv_id":"2407.04295","citing_title":"Jailbreak Attacks and Defenses Against Large Language Models: A Survey","ref_index":76,"is_internal_anchor":false},{"citing_arxiv_id":"2310.08419","citing_title":"Jailbreaking Black Box Large Language Models in Twenty Queries","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2605.05116","citing_title":"On the Hardness of Junking LLMs","ref_index":45,"is_internal_anchor":false},{"citing_arxiv_id":"2605.01899","citing_title":"Disentangling Intent from Role: Adversarial Self-Play for Persona-Invariant Safety Alignment","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2604.10733","citing_title":"Too Nice to Tell the Truth: Quantifying Agreeableness-Driven Sycophancy in Role-Playing Language Models","ref_index":40,"is_internal_anchor":false},{"citing_arxiv_id":"2604.09750","citing_title":"Conflicts Make Large Reasoning Models Vulnerable to Attacks","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2604.14548","citing_title":"VoxSafeBench: Not Just What Is Said, but Who, How, and Where","ref_index":76,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/YDOTUXDKCOJ2VSOKV7ENL4O5QY","json":"https://pith.science/pith/YDOTUXDKCOJ2VSOKV7ENL4O5QY.json","graph_json":"https://pith.science/api/pith-number/YDOTUXDKCOJ2VSOKV7ENL4O5QY/graph.json","events_json":"https://pith.science/api/pith-number/YDOTUXDKCOJ2VSOKV7ENL4O5QY/events.json","paper":"https://pith.science/paper/YDOTUXDK"},"agent_actions":{"view_html":"https://pith.science/pith/YDOTUXDKCOJ2VSOKV7ENL4O5QY","download_json":"https://pith.science/pith/YDOTUXDKCOJ2VSOKV7ENL4O5QY.json","view_paper":"https://pith.science/paper/YDOTUXDK","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2311.03348&json=true","fetch_graph":"https://pith.science/api/pith-number/YDOTUXDKCOJ2VSOKV7ENL4O5QY/graph.json","fetch_events":"https://pith.science/api/pith-number/YDOTUXDKCOJ2VSOKV7ENL4O5QY/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/YDOTUXDKCOJ2VSOKV7ENL4O5QY/action/timestamp_anchor","attest_storage":"https://pith.science/pith/YDOTUXDKCOJ2VSOKV7ENL4O5QY/action/storage_attestation","attest_author":"https://pith.science/pith/YDOTUXDKCOJ2VSOKV7ENL4O5QY/action/author_attestation","sign_citation":"https://pith.science/pith/YDOTUXDKCOJ2VSOKV7ENL4O5QY/action/citation_signature","submit_replication":"https://pith.science/pith/YDOTUXDKCOJ2VSOKV7ENL4O5QY/action/replication_record"}},"created_at":"2026-07-05T07:16:02.521515+00:00","updated_at":"2026-07-05T07:16:02.521515+00:00"}