{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:JDSSWSI7A4THCVMKTIABOCNJBV","short_pith_number":"pith:JDSSWSI7","schema_version":"1.0","canonical_sha256":"48e52b491f072671558a9a001709a90d728ed4760628dc635a135a7a95591e84","source":{"kind":"arxiv","id":"2606.27210","version":1},"attestation_state":"computed","paper":{"title":"Paved with True Intents: Intent-Aware Training Improves LLM Safety Classification Across Training Regimes","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Iustin S\\^irbu, Jeremias Ferrao, Niclas M\\\"uller-Hof, Traian Rebedea, Yftah Ziser","submitted_at":"2026-06-25T16:03:57Z","abstract_excerpt":"We argue that safety classifiers should model user intent as an explicit signal between the prompt and the final label. To study this, we introduce AIMS, a human-annotated dataset of 1,724 difficult safety prompts, each paired with an intent description and harm label. We use AIMS to evaluate intent-aware training across supervised fine-tuning, preference learning, reasoning distillation, and reinforcement learning. Despite its size, AIMS enables competitive safety classifiers across training regimes: DPO from model-generated intent errors improves over SFT, and intent-conditioned distillation"},"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":"2606.27210","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2026-06-25T16:03:57Z","cross_cats_sorted":[],"title_canon_sha256":"5e581e4be37ff266623f5e0a53b906d42e2accceb06a36a24b3397aed4fb39ea","abstract_canon_sha256":"f237c55b554ff9f0f70050af605fd319d5a1f8d90439d4fcf319003d4899588b"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-26T01:16:13.687113Z","signature_b64":"HIKDpiU2edHIvRQm84GQZJQIs8M+Gt/pbCUe5aZGGrXabPUD2aZeVF8ev4hinhb+tFwZfeHV6qN/m0YmeAVjAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"48e52b491f072671558a9a001709a90d728ed4760628dc635a135a7a95591e84","last_reissued_at":"2026-06-26T01:16:13.686726Z","signature_status":"signed_v1","first_computed_at":"2026-06-26T01:16:13.686726Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Paved with True Intents: Intent-Aware Training Improves LLM Safety Classification Across Training Regimes","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Iustin S\\^irbu, Jeremias Ferrao, Niclas M\\\"uller-Hof, Traian Rebedea, Yftah Ziser","submitted_at":"2026-06-25T16:03:57Z","abstract_excerpt":"We argue that safety classifiers should model user intent as an explicit signal between the prompt and the final label. To study this, we introduce AIMS, a human-annotated dataset of 1,724 difficult safety prompts, each paired with an intent description and harm label. We use AIMS to evaluate intent-aware training across supervised fine-tuning, preference learning, reasoning distillation, and reinforcement learning. Despite its size, AIMS enables competitive safety classifiers across training regimes: DPO from model-generated intent errors improves over SFT, and intent-conditioned distillation"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.27210","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/2606.27210/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":"2606.27210","created_at":"2026-06-26T01:16:13.686784+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.27210v1","created_at":"2026-06-26T01:16:13.686784+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.27210","created_at":"2026-06-26T01:16:13.686784+00:00"},{"alias_kind":"pith_short_12","alias_value":"JDSSWSI7A4TH","created_at":"2026-06-26T01:16:13.686784+00:00"},{"alias_kind":"pith_short_16","alias_value":"JDSSWSI7A4THCVMK","created_at":"2026-06-26T01:16:13.686784+00:00"},{"alias_kind":"pith_short_8","alias_value":"JDSSWSI7","created_at":"2026-06-26T01:16:13.686784+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/JDSSWSI7A4THCVMKTIABOCNJBV","json":"https://pith.science/pith/JDSSWSI7A4THCVMKTIABOCNJBV.json","graph_json":"https://pith.science/api/pith-number/JDSSWSI7A4THCVMKTIABOCNJBV/graph.json","events_json":"https://pith.science/api/pith-number/JDSSWSI7A4THCVMKTIABOCNJBV/events.json","paper":"https://pith.science/paper/JDSSWSI7"},"agent_actions":{"view_html":"https://pith.science/pith/JDSSWSI7A4THCVMKTIABOCNJBV","download_json":"https://pith.science/pith/JDSSWSI7A4THCVMKTIABOCNJBV.json","view_paper":"https://pith.science/paper/JDSSWSI7","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.27210&json=true","fetch_graph":"https://pith.science/api/pith-number/JDSSWSI7A4THCVMKTIABOCNJBV/graph.json","fetch_events":"https://pith.science/api/pith-number/JDSSWSI7A4THCVMKTIABOCNJBV/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/JDSSWSI7A4THCVMKTIABOCNJBV/action/timestamp_anchor","attest_storage":"https://pith.science/pith/JDSSWSI7A4THCVMKTIABOCNJBV/action/storage_attestation","attest_author":"https://pith.science/pith/JDSSWSI7A4THCVMKTIABOCNJBV/action/author_attestation","sign_citation":"https://pith.science/pith/JDSSWSI7A4THCVMKTIABOCNJBV/action/citation_signature","submit_replication":"https://pith.science/pith/JDSSWSI7A4THCVMKTIABOCNJBV/action/replication_record"}},"created_at":"2026-06-26T01:16:13.686784+00:00","updated_at":"2026-06-26T01:16:13.686784+00:00"}