{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:2M623PTWVQHC5FSKJDCC5N3I6A","short_pith_number":"pith:2M623PTW","schema_version":"1.0","canonical_sha256":"d33dadbe76ac0e2e964a48c42eb768f007038ad43377143b27d1ee3ad7f17084","source":{"kind":"arxiv","id":"2607.20472","version":1},"attestation_state":"computed","paper":{"title":"Robust Critics: Defending LLMs Against Multi-Turn Attacks","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Arunesh Sinha, Pradeep Varakantham, Roman Belaire","submitted_at":"2026-05-24T05:49:14Z","abstract_excerpt":"When a user asks a language model something harmful, is it a genuine attack or a misunderstood but well-meaning question? This ambiguity is one of the central challenges of LLM safety. A model that assumes the worst harms legitimate users; one that assumes the best is easily exploited. The problem is compounded in multi-turn dialogue, where an attacker's true intent may only reveal itself gradually across many exchanges, yet existing safety frameworks apply a contextual bandit treatment, ignoring the trajectory of the conversation.\n  To that end, we propose Dialogue Critic Guided Sampling (DCG"},"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":"2607.20472","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-24T05:49:14Z","cross_cats_sorted":[],"title_canon_sha256":"15ccc32d25cb3361093224b610e1957551be2829b49ff9811bc2fbd07859d725","abstract_canon_sha256":"9073112c5e28e4eca530635ba7f84738b8a1430a3b7ecf8fcfdb1d3692c4794e"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-24T00:23:17.573107Z","signature_b64":"z3nRdvi2Nm+NPVRYWsX45q+V1r+DK5yLJmLTGLucnoVGOHYOU2AmJvKbnHIgHZx/x6zAlEG5R0IJSrF0QlcXBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"d33dadbe76ac0e2e964a48c42eb768f007038ad43377143b27d1ee3ad7f17084","last_reissued_at":"2026-07-24T00:23:17.572244Z","signature_status":"signed_v1","first_computed_at":"2026-07-24T00:23:17.572244Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Robust Critics: Defending LLMs Against Multi-Turn Attacks","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Arunesh Sinha, Pradeep Varakantham, Roman Belaire","submitted_at":"2026-05-24T05:49:14Z","abstract_excerpt":"When a user asks a language model something harmful, is it a genuine attack or a misunderstood but well-meaning question? This ambiguity is one of the central challenges of LLM safety. A model that assumes the worst harms legitimate users; one that assumes the best is easily exploited. The problem is compounded in multi-turn dialogue, where an attacker's true intent may only reveal itself gradually across many exchanges, yet existing safety frameworks apply a contextual bandit treatment, ignoring the trajectory of the conversation.\n  To that end, we propose Dialogue Critic Guided Sampling (DCG"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2607.20472","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/2607.20472/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":"2607.20472","created_at":"2026-07-24T00:23:17.572684+00:00"},{"alias_kind":"arxiv_version","alias_value":"2607.20472v1","created_at":"2026-07-24T00:23:17.572684+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2607.20472","created_at":"2026-07-24T00:23:17.572684+00:00"},{"alias_kind":"pith_short_12","alias_value":"2M623PTWVQHC","created_at":"2026-07-24T00:23:17.572684+00:00"},{"alias_kind":"pith_short_16","alias_value":"2M623PTWVQHC5FSK","created_at":"2026-07-24T00:23:17.572684+00:00"},{"alias_kind":"pith_short_8","alias_value":"2M623PTW","created_at":"2026-07-24T00:23:17.572684+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/2M623PTWVQHC5FSKJDCC5N3I6A","json":"https://pith.science/pith/2M623PTWVQHC5FSKJDCC5N3I6A.json","graph_json":"https://pith.science/api/pith-number/2M623PTWVQHC5FSKJDCC5N3I6A/graph.json","events_json":"https://pith.science/api/pith-number/2M623PTWVQHC5FSKJDCC5N3I6A/events.json","paper":"https://pith.science/paper/2M623PTW"},"agent_actions":{"view_html":"https://pith.science/pith/2M623PTWVQHC5FSKJDCC5N3I6A","download_json":"https://pith.science/pith/2M623PTWVQHC5FSKJDCC5N3I6A.json","view_paper":"https://pith.science/paper/2M623PTW","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2607.20472&json=true","fetch_graph":"https://pith.science/api/pith-number/2M623PTWVQHC5FSKJDCC5N3I6A/graph.json","fetch_events":"https://pith.science/api/pith-number/2M623PTWVQHC5FSKJDCC5N3I6A/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/2M623PTWVQHC5FSKJDCC5N3I6A/action/timestamp_anchor","attest_storage":"https://pith.science/pith/2M623PTWVQHC5FSKJDCC5N3I6A/action/storage_attestation","attest_author":"https://pith.science/pith/2M623PTWVQHC5FSKJDCC5N3I6A/action/author_attestation","sign_citation":"https://pith.science/pith/2M623PTWVQHC5FSKJDCC5N3I6A/action/citation_signature","submit_replication":"https://pith.science/pith/2M623PTWVQHC5FSKJDCC5N3I6A/action/replication_record"}},"created_at":"2026-07-24T00:23:17.572684+00:00","updated_at":"2026-07-24T00:23:17.572684+00:00"}