{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:PX3B3GGNLJY7YLNTX4AK7MJF7Z","short_pith_number":"pith:PX3B3GGN","schema_version":"1.0","canonical_sha256":"7df61d98cd5a71fc2db3bf00afb125fe688dcd9bf099a5146dfa8531e77b9625","source":{"kind":"arxiv","id":"2507.08284","version":1},"attestation_state":"computed","paper":{"title":"Lightweight Safety Guardrails via Synthetic Data and RL-guided Adversarial Training","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Aleksei Ilin, Gor Matevosyan, Haluk Noyan Tokgozoglu, Muqun Li, Riyaaz Shaik, Suhaa Dada, Vladimir Eremin, Xueying Ma","submitted_at":"2025-07-11T03:17:58Z","abstract_excerpt":"We introduce a lightweight yet highly effective safety guardrail framework for language models, demonstrating that small-scale language models can achieve, and even surpass, the performance of larger counterparts in content moderation tasks. This is accomplished through high-fidelity synthetic data generation and adversarial training. The synthetic data generation process begins with human-curated seed data, which undergoes query augmentation and paraphrasing to create diverse and contextually rich examples. This augmented data is then subjected to multiple rounds of curation, ensuring high fi"},"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":"2507.08284","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-07-11T03:17:58Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"81ed26466b6d273a223c84856471833954cce7ff89d087f7f6e822db5f8e3017","abstract_canon_sha256":"b2f2b0e54170a85572022cd3747b00fbc1c3b74f8d688ef9352a85bee7986188"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:35:20.758289Z","signature_b64":"Z6jwA61rXGonK+W8gPWXHpptksybXyUJbCyj3obPTCZhRBwHpokoglCy6tqqlhgwETFXK9lpIHblTfGvruXmCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"7df61d98cd5a71fc2db3bf00afb125fe688dcd9bf099a5146dfa8531e77b9625","last_reissued_at":"2026-07-05T11:35:20.757802Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:35:20.757802Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Lightweight Safety Guardrails via Synthetic Data and RL-guided Adversarial Training","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Aleksei Ilin, Gor Matevosyan, Haluk Noyan Tokgozoglu, Muqun Li, Riyaaz Shaik, Suhaa Dada, Vladimir Eremin, Xueying Ma","submitted_at":"2025-07-11T03:17:58Z","abstract_excerpt":"We introduce a lightweight yet highly effective safety guardrail framework for language models, demonstrating that small-scale language models can achieve, and even surpass, the performance of larger counterparts in content moderation tasks. This is accomplished through high-fidelity synthetic data generation and adversarial training. The synthetic data generation process begins with human-curated seed data, which undergoes query augmentation and paraphrasing to create diverse and contextually rich examples. This augmented data is then subjected to multiple rounds of curation, ensuring high fi"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2507.08284","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/2507.08284/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":"2507.08284","created_at":"2026-07-05T11:35:20.757862+00:00"},{"alias_kind":"arxiv_version","alias_value":"2507.08284v1","created_at":"2026-07-05T11:35:20.757862+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2507.08284","created_at":"2026-07-05T11:35:20.757862+00:00"},{"alias_kind":"pith_short_12","alias_value":"PX3B3GGNLJY7","created_at":"2026-07-05T11:35:20.757862+00:00"},{"alias_kind":"pith_short_16","alias_value":"PX3B3GGNLJY7YLNT","created_at":"2026-07-05T11:35:20.757862+00:00"},{"alias_kind":"pith_short_8","alias_value":"PX3B3GGN","created_at":"2026-07-05T11:35:20.757862+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/PX3B3GGNLJY7YLNTX4AK7MJF7Z","json":"https://pith.science/pith/PX3B3GGNLJY7YLNTX4AK7MJF7Z.json","graph_json":"https://pith.science/api/pith-number/PX3B3GGNLJY7YLNTX4AK7MJF7Z/graph.json","events_json":"https://pith.science/api/pith-number/PX3B3GGNLJY7YLNTX4AK7MJF7Z/events.json","paper":"https://pith.science/paper/PX3B3GGN"},"agent_actions":{"view_html":"https://pith.science/pith/PX3B3GGNLJY7YLNTX4AK7MJF7Z","download_json":"https://pith.science/pith/PX3B3GGNLJY7YLNTX4AK7MJF7Z.json","view_paper":"https://pith.science/paper/PX3B3GGN","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2507.08284&json=true","fetch_graph":"https://pith.science/api/pith-number/PX3B3GGNLJY7YLNTX4AK7MJF7Z/graph.json","fetch_events":"https://pith.science/api/pith-number/PX3B3GGNLJY7YLNTX4AK7MJF7Z/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/PX3B3GGNLJY7YLNTX4AK7MJF7Z/action/timestamp_anchor","attest_storage":"https://pith.science/pith/PX3B3GGNLJY7YLNTX4AK7MJF7Z/action/storage_attestation","attest_author":"https://pith.science/pith/PX3B3GGNLJY7YLNTX4AK7MJF7Z/action/author_attestation","sign_citation":"https://pith.science/pith/PX3B3GGNLJY7YLNTX4AK7MJF7Z/action/citation_signature","submit_replication":"https://pith.science/pith/PX3B3GGNLJY7YLNTX4AK7MJF7Z/action/replication_record"}},"created_at":"2026-07-05T11:35:20.757862+00:00","updated_at":"2026-07-05T11:35:20.757862+00:00"}