{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:G5RBYGHAV3EVENHTH2LFBGBMU5","short_pith_number":"pith:G5RBYGHA","schema_version":"1.0","canonical_sha256":"37621c18e0aec95234f33e9650982ca75785e259c351e4c8f2b520122bac73cf","source":{"kind":"arxiv","id":"2507.15328","version":1},"attestation_state":"computed","paper":{"title":"On the Inevitability of Left-Leaning Political Bias in Aligned Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CY"],"primary_cat":"cs.CL","authors_text":"Thilo Hagendorff","submitted_at":"2025-07-21T07:37:28Z","abstract_excerpt":"The guiding principle of AI alignment is to train large language models (LLMs) to be harmless, helpful, and honest (HHH). At the same time, there are mounting concerns that LLMs exhibit a left-wing political bias. Yet, the commitment to AI alignment cannot be harmonized with the latter critique. In this article, I argue that intelligent systems that are trained to be harmless and honest must necessarily exhibit left-wing political bias. Normative assumptions underlying alignment objectives inherently concur with progressive moral frameworks and left-wing principles, emphasizing harm avoidance,"},"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":"2507.15328","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-07-21T07:37:28Z","cross_cats_sorted":["cs.CY"],"title_canon_sha256":"fbe41ae609b0127ba1533f49480e766affd591d4333870029c7580bd95b46e72","abstract_canon_sha256":"113503147184c23124710c1674e5a68da5c64033f21ab04d3bcef1144cef70d1"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:40:22.783109Z","signature_b64":"sS04OVyiSQE21FYSy/cM19RWt+Z3iKgeT/PysOzmS3zjT07UCXagvQ1Bt2lpVuiVZq8RDEHr4WW6uYzJsxuHBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"37621c18e0aec95234f33e9650982ca75785e259c351e4c8f2b520122bac73cf","last_reissued_at":"2026-07-05T11:40:22.782382Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:40:22.782382Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"On the Inevitability of Left-Leaning Political Bias in Aligned Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CY"],"primary_cat":"cs.CL","authors_text":"Thilo Hagendorff","submitted_at":"2025-07-21T07:37:28Z","abstract_excerpt":"The guiding principle of AI alignment is to train large language models (LLMs) to be harmless, helpful, and honest (HHH). At the same time, there are mounting concerns that LLMs exhibit a left-wing political bias. Yet, the commitment to AI alignment cannot be harmonized with the latter critique. In this article, I argue that intelligent systems that are trained to be harmless and honest must necessarily exhibit left-wing political bias. Normative assumptions underlying alignment objectives inherently concur with progressive moral frameworks and left-wing principles, emphasizing harm avoidance,"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2507.15328","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/2507.15328/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":"2507.15328","created_at":"2026-07-05T11:40:22.782475+00:00"},{"alias_kind":"arxiv_version","alias_value":"2507.15328v1","created_at":"2026-07-05T11:40:22.782475+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2507.15328","created_at":"2026-07-05T11:40:22.782475+00:00"},{"alias_kind":"pith_short_12","alias_value":"G5RBYGHAV3EV","created_at":"2026-07-05T11:40:22.782475+00:00"},{"alias_kind":"pith_short_16","alias_value":"G5RBYGHAV3EVENHT","created_at":"2026-07-05T11:40:22.782475+00:00"},{"alias_kind":"pith_short_8","alias_value":"G5RBYGHA","created_at":"2026-07-05T11:40:22.782475+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":3,"sample":[{"citing_arxiv_id":"2606.13111","citing_title":"M\\\"OVE: A Holistic LLM Benchmark for the German Public Sector","ref_index":127,"is_internal_anchor":true},{"citing_arxiv_id":"2606.12426","citing_title":"Two Wrongs, No Right: Auditing Social-Desirability Bias in LLM Annotators for Computational Social Science","ref_index":5,"is_internal_anchor":true},{"citing_arxiv_id":"2509.22367","citing_title":"What Is The Political Content in LLMs' Pre- and Post-Training Data?","ref_index":16,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/G5RBYGHAV3EVENHTH2LFBGBMU5","json":"https://pith.science/pith/G5RBYGHAV3EVENHTH2LFBGBMU5.json","graph_json":"https://pith.science/api/pith-number/G5RBYGHAV3EVENHTH2LFBGBMU5/graph.json","events_json":"https://pith.science/api/pith-number/G5RBYGHAV3EVENHTH2LFBGBMU5/events.json","paper":"https://pith.science/paper/G5RBYGHA"},"agent_actions":{"view_html":"https://pith.science/pith/G5RBYGHAV3EVENHTH2LFBGBMU5","download_json":"https://pith.science/pith/G5RBYGHAV3EVENHTH2LFBGBMU5.json","view_paper":"https://pith.science/paper/G5RBYGHA","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2507.15328&json=true","fetch_graph":"https://pith.science/api/pith-number/G5RBYGHAV3EVENHTH2LFBGBMU5/graph.json","fetch_events":"https://pith.science/api/pith-number/G5RBYGHAV3EVENHTH2LFBGBMU5/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/G5RBYGHAV3EVENHTH2LFBGBMU5/action/timestamp_anchor","attest_storage":"https://pith.science/pith/G5RBYGHAV3EVENHTH2LFBGBMU5/action/storage_attestation","attest_author":"https://pith.science/pith/G5RBYGHAV3EVENHTH2LFBGBMU5/action/author_attestation","sign_citation":"https://pith.science/pith/G5RBYGHAV3EVENHTH2LFBGBMU5/action/citation_signature","submit_replication":"https://pith.science/pith/G5RBYGHAV3EVENHTH2LFBGBMU5/action/replication_record"}},"created_at":"2026-07-05T11:40:22.782475+00:00","updated_at":"2026-07-05T11:40:22.782475+00:00"}