{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:X4NSOE73KOYAVIX2KCNUQBQNPM","short_pith_number":"pith:X4NSOE73","schema_version":"1.0","canonical_sha256":"bf1b2713fb53b00aa2fa509b48060d7b16403f9fe7cc9ff46a754a629ead24f2","source":{"kind":"arxiv","id":"2505.10838","version":1},"attestation_state":"computed","paper":{"title":"LARGO: Latent Adversarial Reflection through Gradient Optimization for Jailbreaking LLMs","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL","cs.CR"],"primary_cat":"cs.LG","authors_text":"Chengzhi Mao, Hao Wang, Ran Li","submitted_at":"2025-05-16T04:12:16Z","abstract_excerpt":"Efficient red-teaming method to uncover vulnerabilities in Large Language Models (LLMs) is crucial. While recent attacks often use LLMs as optimizers, the discrete language space make gradient-based methods struggle. We introduce LARGO (Latent Adversarial Reflection through Gradient Optimization), a novel latent self-reflection attack that reasserts the power of gradient-based optimization for generating fluent jailbreaking prompts. By operating within the LLM's continuous latent space, LARGO first optimizes an adversarial latent vector and then recursively call the same LLM to decode the late"},"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":"2505.10838","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-05-16T04:12:16Z","cross_cats_sorted":["cs.CL","cs.CR"],"title_canon_sha256":"e57143a4dc5676825a7b62ad43fd20190ba84b6ad98dd735f923e96491a5db28","abstract_canon_sha256":"69f3182b68c675a63343271a11a1cb478f3d3c19cb9f54a82ba416768280c1af"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:04:06.909028Z","signature_b64":"vhHZ9blPeImyz0OzkK0wPI149H5Jz1o2LqoE8HN7kQ3vn25YOCJF0ojJqS9gVZzfDMINaBn5vLOOGyfWsVhlDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"bf1b2713fb53b00aa2fa509b48060d7b16403f9fe7cc9ff46a754a629ead24f2","last_reissued_at":"2026-07-05T11:04:06.908404Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:04:06.908404Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"LARGO: Latent Adversarial Reflection through Gradient Optimization for Jailbreaking LLMs","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL","cs.CR"],"primary_cat":"cs.LG","authors_text":"Chengzhi Mao, Hao Wang, Ran Li","submitted_at":"2025-05-16T04:12:16Z","abstract_excerpt":"Efficient red-teaming method to uncover vulnerabilities in Large Language Models (LLMs) is crucial. While recent attacks often use LLMs as optimizers, the discrete language space make gradient-based methods struggle. We introduce LARGO (Latent Adversarial Reflection through Gradient Optimization), a novel latent self-reflection attack that reasserts the power of gradient-based optimization for generating fluent jailbreaking prompts. By operating within the LLM's continuous latent space, LARGO first optimizes an adversarial latent vector and then recursively call the same LLM to decode the late"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.10838","kind":"arxiv","version":1},"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/2505.10838/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":"2505.10838","created_at":"2026-07-05T11:04:06.908470+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.10838v1","created_at":"2026-07-05T11:04:06.908470+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.10838","created_at":"2026-07-05T11:04:06.908470+00:00"},{"alias_kind":"pith_short_12","alias_value":"X4NSOE73KOYA","created_at":"2026-07-05T11:04:06.908470+00:00"},{"alias_kind":"pith_short_16","alias_value":"X4NSOE73KOYAVIX2","created_at":"2026-07-05T11:04:06.908470+00:00"},{"alias_kind":"pith_short_8","alias_value":"X4NSOE73","created_at":"2026-07-05T11:04:06.908470+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.21362","citing_title":"LASH: Adaptive Semantic Hybridization for Black-Box Jailbreaking of Large Language Models","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12813","citing_title":"REALISTA: Realistic Latent Adversarial Attacks that Elicit LLM Hallucinations","ref_index":75,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/X4NSOE73KOYAVIX2KCNUQBQNPM","json":"https://pith.science/pith/X4NSOE73KOYAVIX2KCNUQBQNPM.json","graph_json":"https://pith.science/api/pith-number/X4NSOE73KOYAVIX2KCNUQBQNPM/graph.json","events_json":"https://pith.science/api/pith-number/X4NSOE73KOYAVIX2KCNUQBQNPM/events.json","paper":"https://pith.science/paper/X4NSOE73"},"agent_actions":{"view_html":"https://pith.science/pith/X4NSOE73KOYAVIX2KCNUQBQNPM","download_json":"https://pith.science/pith/X4NSOE73KOYAVIX2KCNUQBQNPM.json","view_paper":"https://pith.science/paper/X4NSOE73","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.10838&json=true","fetch_graph":"https://pith.science/api/pith-number/X4NSOE73KOYAVIX2KCNUQBQNPM/graph.json","fetch_events":"https://pith.science/api/pith-number/X4NSOE73KOYAVIX2KCNUQBQNPM/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/X4NSOE73KOYAVIX2KCNUQBQNPM/action/timestamp_anchor","attest_storage":"https://pith.science/pith/X4NSOE73KOYAVIX2KCNUQBQNPM/action/storage_attestation","attest_author":"https://pith.science/pith/X4NSOE73KOYAVIX2KCNUQBQNPM/action/author_attestation","sign_citation":"https://pith.science/pith/X4NSOE73KOYAVIX2KCNUQBQNPM/action/citation_signature","submit_replication":"https://pith.science/pith/X4NSOE73KOYAVIX2KCNUQBQNPM/action/replication_record"}},"created_at":"2026-07-05T11:04:06.908470+00:00","updated_at":"2026-07-05T11:04:06.908470+00:00"}