{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:JOOMBDGL6C26FDPSH3SCJ5NOSH","short_pith_number":"pith:JOOMBDGL","schema_version":"1.0","canonical_sha256":"4b9cc08ccbf0b5e28df23ee424f5ae91db86ac1201c20ea053500ee802815cb7","source":{"kind":"arxiv","id":"2606.12342","version":1},"attestation_state":"computed","paper":{"title":"ALIGNBEAM : Inference-Time Alignment Transfer via Cross-Vocabulary Logit Mixing","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.ET","cs.LG"],"primary_cat":"cs.CL","authors_text":"Chirag Chawla, Pratinav Seth, Vinay Kumar Sankarapu","submitted_at":"2026-06-10T17:15:28Z","abstract_excerpt":"Domain fine-tuning degrades the safety of large language models: fine-tuned specialists readily comply with harmful prompts framed in domain language. Existing inference-time defenses that mix logits from a safe anchor model require both models to share a vocabulary, which rules them out for the cross-family specialists where safety is most degraded. We present ALIGNBEAM, a training-free method that lifts this restriction by translating anchor logits into the target model's vocabulary token-by-token at each decoding step; a small LLM judge then selects the safest among K candidate continuation"},"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":"2606.12342","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2026-06-10T17:15:28Z","cross_cats_sorted":["cs.AI","cs.ET","cs.LG"],"title_canon_sha256":"941244fdb3d60fc3f7e16a74c738e24fd3104f2fb4e2cc965279725dacea16ef","abstract_canon_sha256":"0a77485743a485e510c3226bbb5033013917fb3599ef01abf43371322cb5c665"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-11T01:11:02.710246Z","signature_b64":"fXQt4jdziMz2277LSc1p56TS+uoK9f+jO9gpVYYdPNza3b+/LohZIOYFjr9J4pidzdoAyjMcq0aI0mFM8i9WAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"4b9cc08ccbf0b5e28df23ee424f5ae91db86ac1201c20ea053500ee802815cb7","last_reissued_at":"2026-06-11T01:11:02.709552Z","signature_status":"signed_v1","first_computed_at":"2026-06-11T01:11:02.709552Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"ALIGNBEAM : Inference-Time Alignment Transfer via Cross-Vocabulary Logit Mixing","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.ET","cs.LG"],"primary_cat":"cs.CL","authors_text":"Chirag Chawla, Pratinav Seth, Vinay Kumar Sankarapu","submitted_at":"2026-06-10T17:15:28Z","abstract_excerpt":"Domain fine-tuning degrades the safety of large language models: fine-tuned specialists readily comply with harmful prompts framed in domain language. Existing inference-time defenses that mix logits from a safe anchor model require both models to share a vocabulary, which rules them out for the cross-family specialists where safety is most degraded. We present ALIGNBEAM, a training-free method that lifts this restriction by translating anchor logits into the target model's vocabulary token-by-token at each decoding step; a small LLM judge then selects the safest among K candidate continuation"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.12342","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/2606.12342/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":"2606.12342","created_at":"2026-06-11T01:11:02.709654+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.12342v1","created_at":"2026-06-11T01:11:02.709654+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.12342","created_at":"2026-06-11T01:11:02.709654+00:00"},{"alias_kind":"pith_short_12","alias_value":"JOOMBDGL6C26","created_at":"2026-06-11T01:11:02.709654+00:00"},{"alias_kind":"pith_short_16","alias_value":"JOOMBDGL6C26FDPS","created_at":"2026-06-11T01:11:02.709654+00:00"},{"alias_kind":"pith_short_8","alias_value":"JOOMBDGL","created_at":"2026-06-11T01:11:02.709654+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/JOOMBDGL6C26FDPSH3SCJ5NOSH","json":"https://pith.science/pith/JOOMBDGL6C26FDPSH3SCJ5NOSH.json","graph_json":"https://pith.science/api/pith-number/JOOMBDGL6C26FDPSH3SCJ5NOSH/graph.json","events_json":"https://pith.science/api/pith-number/JOOMBDGL6C26FDPSH3SCJ5NOSH/events.json","paper":"https://pith.science/paper/JOOMBDGL"},"agent_actions":{"view_html":"https://pith.science/pith/JOOMBDGL6C26FDPSH3SCJ5NOSH","download_json":"https://pith.science/pith/JOOMBDGL6C26FDPSH3SCJ5NOSH.json","view_paper":"https://pith.science/paper/JOOMBDGL","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.12342&json=true","fetch_graph":"https://pith.science/api/pith-number/JOOMBDGL6C26FDPSH3SCJ5NOSH/graph.json","fetch_events":"https://pith.science/api/pith-number/JOOMBDGL6C26FDPSH3SCJ5NOSH/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/JOOMBDGL6C26FDPSH3SCJ5NOSH/action/timestamp_anchor","attest_storage":"https://pith.science/pith/JOOMBDGL6C26FDPSH3SCJ5NOSH/action/storage_attestation","attest_author":"https://pith.science/pith/JOOMBDGL6C26FDPSH3SCJ5NOSH/action/author_attestation","sign_citation":"https://pith.science/pith/JOOMBDGL6C26FDPSH3SCJ5NOSH/action/citation_signature","submit_replication":"https://pith.science/pith/JOOMBDGL6C26FDPSH3SCJ5NOSH/action/replication_record"}},"created_at":"2026-06-11T01:11:02.709654+00:00","updated_at":"2026-06-11T01:11:02.709654+00:00"}