{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:3BXSRL7NP6V4TC4VTAUQXHFTV5","short_pith_number":"pith:3BXSRL7N","schema_version":"1.0","canonical_sha256":"d86f28afed7fabc98b9598290b9cb3af4805a39973c7424243332e7346d31a9f","source":{"kind":"arxiv","id":"2407.16318","version":1},"attestation_state":"computed","paper":{"title":"PrimeGuard: Safe and Helpful LLMs through Tuning-Free Routing","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL","cs.CR","cs.SE"],"primary_cat":"cs.AI","authors_text":"Blazej Manczak, Eliott Zemour, Eric Lin, Vaikkunth Mugunthan","submitted_at":"2024-07-23T09:14:27Z","abstract_excerpt":"Deploying language models (LMs) necessitates outputs to be both high-quality and compliant with safety guidelines. Although Inference-Time Guardrails (ITG) offer solutions that shift model output distributions towards compliance, we find that current methods struggle in balancing safety with helpfulness. ITG Methods that safely address non-compliant queries exhibit lower helpfulness while those that prioritize helpfulness compromise on safety. We refer to this trade-off as the guardrail tax, analogous to the alignment tax. To address this, we propose PrimeGuard, a novel ITG method that utilize"},"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":"2407.16318","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2024-07-23T09:14:27Z","cross_cats_sorted":["cs.CL","cs.CR","cs.SE"],"title_canon_sha256":"1ef6b7af1e93bad060df4c557f27d8532794008c251444f1a72caef10f05c172","abstract_canon_sha256":"e32ee27bbd1e6b25ca37c9041433ead9bc3da6b0b2bdfdf7dbcdc69000188d62"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:47:31.364124Z","signature_b64":"13kTmvyZp4b9u++Cxj/BiyuFnQtW9tSKckXX7q4Q9yY+yLSlVjgyj+I6lma/7T+VX+jv15k1fv6haFrVZ5xBAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"d86f28afed7fabc98b9598290b9cb3af4805a39973c7424243332e7346d31a9f","last_reissued_at":"2026-07-05T08:47:31.363639Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:47:31.363639Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"PrimeGuard: Safe and Helpful LLMs through Tuning-Free Routing","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL","cs.CR","cs.SE"],"primary_cat":"cs.AI","authors_text":"Blazej Manczak, Eliott Zemour, Eric Lin, Vaikkunth Mugunthan","submitted_at":"2024-07-23T09:14:27Z","abstract_excerpt":"Deploying language models (LMs) necessitates outputs to be both high-quality and compliant with safety guidelines. Although Inference-Time Guardrails (ITG) offer solutions that shift model output distributions towards compliance, we find that current methods struggle in balancing safety with helpfulness. ITG Methods that safely address non-compliant queries exhibit lower helpfulness while those that prioritize helpfulness compromise on safety. We refer to this trade-off as the guardrail tax, analogous to the alignment tax. To address this, we propose PrimeGuard, a novel ITG method that utilize"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2407.16318","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/2407.16318/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":"2407.16318","created_at":"2026-07-05T08:47:31.363694+00:00"},{"alias_kind":"arxiv_version","alias_value":"2407.16318v1","created_at":"2026-07-05T08:47:31.363694+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2407.16318","created_at":"2026-07-05T08:47:31.363694+00:00"},{"alias_kind":"pith_short_12","alias_value":"3BXSRL7NP6V4","created_at":"2026-07-05T08:47:31.363694+00:00"},{"alias_kind":"pith_short_16","alias_value":"3BXSRL7NP6V4TC4V","created_at":"2026-07-05T08:47:31.363694+00:00"},{"alias_kind":"pith_short_8","alias_value":"3BXSRL7N","created_at":"2026-07-05T08:47:31.363694+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2604.18756","citing_title":"Towards Understanding the Robustness of Sparse Autoencoders","ref_index":33,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/3BXSRL7NP6V4TC4VTAUQXHFTV5","json":"https://pith.science/pith/3BXSRL7NP6V4TC4VTAUQXHFTV5.json","graph_json":"https://pith.science/api/pith-number/3BXSRL7NP6V4TC4VTAUQXHFTV5/graph.json","events_json":"https://pith.science/api/pith-number/3BXSRL7NP6V4TC4VTAUQXHFTV5/events.json","paper":"https://pith.science/paper/3BXSRL7N"},"agent_actions":{"view_html":"https://pith.science/pith/3BXSRL7NP6V4TC4VTAUQXHFTV5","download_json":"https://pith.science/pith/3BXSRL7NP6V4TC4VTAUQXHFTV5.json","view_paper":"https://pith.science/paper/3BXSRL7N","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2407.16318&json=true","fetch_graph":"https://pith.science/api/pith-number/3BXSRL7NP6V4TC4VTAUQXHFTV5/graph.json","fetch_events":"https://pith.science/api/pith-number/3BXSRL7NP6V4TC4VTAUQXHFTV5/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/3BXSRL7NP6V4TC4VTAUQXHFTV5/action/timestamp_anchor","attest_storage":"https://pith.science/pith/3BXSRL7NP6V4TC4VTAUQXHFTV5/action/storage_attestation","attest_author":"https://pith.science/pith/3BXSRL7NP6V4TC4VTAUQXHFTV5/action/author_attestation","sign_citation":"https://pith.science/pith/3BXSRL7NP6V4TC4VTAUQXHFTV5/action/citation_signature","submit_replication":"https://pith.science/pith/3BXSRL7NP6V4TC4VTAUQXHFTV5/action/replication_record"}},"created_at":"2026-07-05T08:47:31.363694+00:00","updated_at":"2026-07-05T08:47:31.363694+00:00"}