{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:C44PCWVMADHSUO3LNN5V52GKD6","short_pith_number":"pith:C44PCWVM","schema_version":"1.0","canonical_sha256":"1738f15aac00cf2a3b6b6b7b5ee8ca1fb41e11aacf7fe8a8b94432f1f2f9a670","source":{"kind":"arxiv","id":"2601.04389","version":3},"attestation_state":"computed","paper":{"title":"Safety Is Not Universal: The Selective Safety Trap in LLM Alignment","license":"http://creativecommons.org/licenses/by/4.0/","headline":"LLM safety alignment protects some demographic groups far more than others, with defense rates varying up to 42 percent within one model.","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Arlindo Rodrigues Galv\\~ao Filho, Diogo Fernandes Costa Silva, Iago Alves Brito, Julia Soares Dollis, Walcy Santos Rezende Rios","submitted_at":"2026-01-07T20:53:18Z","abstract_excerpt":"Current safety evaluations of large language models (LLMs) create a dangerous illusion of universal protection by aggregating harms under generic categories such as \"Identity Hate\", obscuring vulnerabilities toward specific populations. In this work, we expose the Selective Safety Trap: a systemic failure mode where models robustly defend specific populations while leaving underrepresented communities highly vulnerable to identical adversarial attacks. To systematically audit this phenomenon, we introduce MiJaBench, a bilingual (English-Portuguese) adversarial benchmark comprising 43,961 contr"},"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":"2601.04389","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-01-07T20:53:18Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"cc28d64672e15f525a89530e71aa2adb9c50a1d166cde7c89ec0ff0bf76754b6","abstract_canon_sha256":"1e5858e07c56143bc8ec3894c094d9d5073d73193f6f9129aec9b301cb525788"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-23T02:13:19.865273Z","signature_b64":"UUKO9vKRBVRYjXAJbYJ32zb/0SM8ucD2+D46gNh0FZlnWO5L3yRgQZ2ebZ2AkTdNA/FBE05s3jYLcCqJRlRLAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"1738f15aac00cf2a3b6b6b7b5ee8ca1fb41e11aacf7fe8a8b94432f1f2f9a670","last_reissued_at":"2026-06-23T02:13:19.864820Z","signature_status":"signed_v1","first_computed_at":"2026-06-23T02:13:19.864820Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Safety Is Not Universal: The Selective Safety Trap in LLM Alignment","license":"http://creativecommons.org/licenses/by/4.0/","headline":"LLM safety alignment protects some demographic groups far more than others, with defense rates varying up to 42 percent within one model.","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Arlindo Rodrigues Galv\\~ao Filho, Diogo Fernandes Costa Silva, Iago Alves Brito, Julia Soares Dollis, Walcy Santos Rezende Rios","submitted_at":"2026-01-07T20:53:18Z","abstract_excerpt":"Current safety evaluations of large language models (LLMs) create a dangerous illusion of universal protection by aggregating harms under generic categories such as \"Identity Hate\", obscuring vulnerabilities toward specific populations. In this work, we expose the Selective Safety Trap: a systemic failure mode where models robustly defend specific populations while leaving underrepresented communities highly vulnerable to identical adversarial attacks. To systematically audit this phenomenon, we introduce MiJaBench, a bilingual (English-Portuguese) adversarial benchmark comprising 43,961 contr"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"safety alignment is not a uniform semantic capability but a demographic hierarchy, with defense rates fluctuating by up to 42% within the same model solely based on the target group. This disparity persists across architectures and languages and is amplified by scaling, indicating that current alignment methods learn group-specific safeguards rather than a generalized notion of harm.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The 43,961 prompts in MiJaBench represent equivalent adversarial difficulty and comparable harm potential across the 16 demographic groups; if prompt construction or translation introduces systematic differences in attack strength, the measured defense-rate gaps would be artifacts rather than evidence of selective safety.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Safety alignment in LLMs is not uniform but forms a demographic hierarchy, with defense rates varying by up to 42% across groups; a new benchmark and DPO method demonstrate transferable safety.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"LLM safety alignment protects some demographic groups far more than others, with defense rates varying up to 42 percent within one model.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"778f161bad109f5978dfec63c80b3db8de53963f3a35b057264396edae4cfb6e"},"source":{"id":"2601.04389","kind":"arxiv","version":3},"verdict":{"id":"968a7583-bbd2-48a4-9cd5-090c0dbbb82b","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-16T15:59:14.537869Z","strongest_claim":"safety alignment is not a uniform semantic capability but a demographic hierarchy, with defense rates fluctuating by up to 42% within the same model solely based on the target group. This disparity persists across architectures and languages and is amplified by scaling, indicating that current alignment methods learn group-specific safeguards rather than a generalized notion of harm.","one_line_summary":"Safety alignment in LLMs is not uniform but forms a demographic hierarchy, with defense rates varying by up to 42% across groups; a new benchmark and DPO method demonstrate transferable safety.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The 43,961 prompts in MiJaBench represent equivalent adversarial difficulty and comparable harm potential across the 16 demographic groups; if prompt construction or translation introduces systematic differences in attack strength, the measured defense-rate gaps would be artifacts rather than evidence of selective safety.","pith_extraction_headline":"LLM safety alignment protects some demographic groups far more than others, with defense rates varying up to 42 percent within one model."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2601.04389/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":"2601.04389","created_at":"2026-06-23T02:13:19.864876+00:00"},{"alias_kind":"arxiv_version","alias_value":"2601.04389v3","created_at":"2026-06-23T02:13:19.864876+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2601.04389","created_at":"2026-06-23T02:13:19.864876+00:00"},{"alias_kind":"pith_short_12","alias_value":"C44PCWVMADHS","created_at":"2026-06-23T02:13:19.864876+00:00"},{"alias_kind":"pith_short_16","alias_value":"C44PCWVMADHSUO3L","created_at":"2026-06-23T02:13:19.864876+00:00"},{"alias_kind":"pith_short_8","alias_value":"C44PCWVM","created_at":"2026-06-23T02:13:19.864876+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/C44PCWVMADHSUO3LNN5V52GKD6","json":"https://pith.science/pith/C44PCWVMADHSUO3LNN5V52GKD6.json","graph_json":"https://pith.science/api/pith-number/C44PCWVMADHSUO3LNN5V52GKD6/graph.json","events_json":"https://pith.science/api/pith-number/C44PCWVMADHSUO3LNN5V52GKD6/events.json","paper":"https://pith.science/paper/C44PCWVM"},"agent_actions":{"view_html":"https://pith.science/pith/C44PCWVMADHSUO3LNN5V52GKD6","download_json":"https://pith.science/pith/C44PCWVMADHSUO3LNN5V52GKD6.json","view_paper":"https://pith.science/paper/C44PCWVM","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2601.04389&json=true","fetch_graph":"https://pith.science/api/pith-number/C44PCWVMADHSUO3LNN5V52GKD6/graph.json","fetch_events":"https://pith.science/api/pith-number/C44PCWVMADHSUO3LNN5V52GKD6/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/C44PCWVMADHSUO3LNN5V52GKD6/action/timestamp_anchor","attest_storage":"https://pith.science/pith/C44PCWVMADHSUO3LNN5V52GKD6/action/storage_attestation","attest_author":"https://pith.science/pith/C44PCWVMADHSUO3LNN5V52GKD6/action/author_attestation","sign_citation":"https://pith.science/pith/C44PCWVMADHSUO3LNN5V52GKD6/action/citation_signature","submit_replication":"https://pith.science/pith/C44PCWVMADHSUO3LNN5V52GKD6/action/replication_record"}},"created_at":"2026-06-23T02:13:19.864876+00:00","updated_at":"2026-06-23T02:13:19.864876+00:00"}