{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:4553VIKCWOU6GQ3KUDPULHFIQG","short_pith_number":"pith:4553VIKC","schema_version":"1.0","canonical_sha256":"e77bbaa142b3a9e3436aa0df459ca881ae636e0d0b211bbfcc01e073832f63bf","source":{"kind":"arxiv","id":"2507.14987","version":1},"attestation_state":"computed","paper":{"title":"AlphaAlign: Incentivizing Safety Alignment with Extremely Simplified Reinforcement Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CR","cs.LG"],"primary_cat":"cs.AI","authors_text":"An Zhang, Leheng Sheng, Xiang Wang, XiuYu Zhang, Yi Zhang, Yuxin Chen, Zhenkai Liang","submitted_at":"2025-07-20T14:47:03Z","abstract_excerpt":"Large language models (LLMs), despite possessing latent safety understanding from their vast pretraining data, remain vulnerable to generating harmful content and exhibit issues such as over-refusal and utility degradation after safety alignment. Current safety alignment methods often result in superficial refusal shortcuts or rely on intensive supervision for reasoning-based approaches, failing to fully leverage the model's intrinsic safety self-awareness. We propose \\textbf{AlphaAlign}, a simple yet effective pure reinforcement learning (RL) framework with verifiable safety reward designed t"},"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":"2507.14987","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.AI","submitted_at":"2025-07-20T14:47:03Z","cross_cats_sorted":["cs.CR","cs.LG"],"title_canon_sha256":"0b33d13080add23998e147b2632c03e68b6573e12593eae70536630ae0c8ab5f","abstract_canon_sha256":"ee2dc46a48737d881c77c25ea2cbf18b3f9523b86d8bdb13173120911df8ae92"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:40:16.284776Z","signature_b64":"RRgb37Dv9JrsKXuJ4Ksda9Bt+bcAH+yOrjOZWAD0/mWHpAuQfYljgP1TWG6WQYYE4SGx6NRQxPy+ephKsOK5BQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"e77bbaa142b3a9e3436aa0df459ca881ae636e0d0b211bbfcc01e073832f63bf","last_reissued_at":"2026-07-05T11:40:16.284280Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:40:16.284280Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"AlphaAlign: Incentivizing Safety Alignment with Extremely Simplified Reinforcement Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CR","cs.LG"],"primary_cat":"cs.AI","authors_text":"An Zhang, Leheng Sheng, Xiang Wang, XiuYu Zhang, Yi Zhang, Yuxin Chen, Zhenkai Liang","submitted_at":"2025-07-20T14:47:03Z","abstract_excerpt":"Large language models (LLMs), despite possessing latent safety understanding from their vast pretraining data, remain vulnerable to generating harmful content and exhibit issues such as over-refusal and utility degradation after safety alignment. Current safety alignment methods often result in superficial refusal shortcuts or rely on intensive supervision for reasoning-based approaches, failing to fully leverage the model's intrinsic safety self-awareness. We propose \\textbf{AlphaAlign}, a simple yet effective pure reinforcement learning (RL) framework with verifiable safety reward designed t"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2507.14987","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/2507.14987/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":"2507.14987","created_at":"2026-07-05T11:40:16.284342+00:00"},{"alias_kind":"arxiv_version","alias_value":"2507.14987v1","created_at":"2026-07-05T11:40:16.284342+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2507.14987","created_at":"2026-07-05T11:40:16.284342+00:00"},{"alias_kind":"pith_short_12","alias_value":"4553VIKCWOU6","created_at":"2026-07-05T11:40:16.284342+00:00"},{"alias_kind":"pith_short_16","alias_value":"4553VIKCWOU6GQ3K","created_at":"2026-07-05T11:40:16.284342+00:00"},{"alias_kind":"pith_short_8","alias_value":"4553VIKC","created_at":"2026-07-05T11:40:16.284342+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.25442","citing_title":"PolicyAlign: Direct Policy-Based Safety Alignment for Large Language Models","ref_index":41,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08936","citing_title":"Self-ReSET: Learning to Self-Recover from Unsafe Reasoning Trajectories","ref_index":33,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/4553VIKCWOU6GQ3KUDPULHFIQG","json":"https://pith.science/pith/4553VIKCWOU6GQ3KUDPULHFIQG.json","graph_json":"https://pith.science/api/pith-number/4553VIKCWOU6GQ3KUDPULHFIQG/graph.json","events_json":"https://pith.science/api/pith-number/4553VIKCWOU6GQ3KUDPULHFIQG/events.json","paper":"https://pith.science/paper/4553VIKC"},"agent_actions":{"view_html":"https://pith.science/pith/4553VIKCWOU6GQ3KUDPULHFIQG","download_json":"https://pith.science/pith/4553VIKCWOU6GQ3KUDPULHFIQG.json","view_paper":"https://pith.science/paper/4553VIKC","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2507.14987&json=true","fetch_graph":"https://pith.science/api/pith-number/4553VIKCWOU6GQ3KUDPULHFIQG/graph.json","fetch_events":"https://pith.science/api/pith-number/4553VIKCWOU6GQ3KUDPULHFIQG/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/4553VIKCWOU6GQ3KUDPULHFIQG/action/timestamp_anchor","attest_storage":"https://pith.science/pith/4553VIKCWOU6GQ3KUDPULHFIQG/action/storage_attestation","attest_author":"https://pith.science/pith/4553VIKCWOU6GQ3KUDPULHFIQG/action/author_attestation","sign_citation":"https://pith.science/pith/4553VIKCWOU6GQ3KUDPULHFIQG/action/citation_signature","submit_replication":"https://pith.science/pith/4553VIKCWOU6GQ3KUDPULHFIQG/action/replication_record"}},"created_at":"2026-07-05T11:40:16.284342+00:00","updated_at":"2026-07-05T11:40:16.284342+00:00"}