{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:EUKSSP33GM22D4UGDYOF5TRNSC","short_pith_number":"pith:EUKSSP33","schema_version":"1.0","canonical_sha256":"2515293f7b3335a1f2861e1c5ece2d908abae8f02b3fd4a4397d1a0a8b0ba9aa","source":{"kind":"arxiv","id":"2510.12229","version":3},"attestation_state":"computed","paper":{"title":"Analysing Moral Bias in Finetuned LLMs through Mechanistic Interpretability","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Bianca Raimondi, Daniela Dalbagno, Maurizio Gabbrielli","submitted_at":"2025-10-14T07:31:29Z","abstract_excerpt":"Large language models (LLMs) have been shown to internalize human-like biases during finetuning, yet the mechanisms by which these biases manifest remain unclear. In this work, we investigated whether the well-known Knobe effect, a moral bias in intentionality judgements, emerges in finetuned LLMs and whether it can be traced back to specific components of the model. We conducted a Layer-Patching analysis across 3 open-weights LLMs and demonstrated that the bias is not only learned during finetuning but also localized in a specific set of layers. Surprisingly, we found that patching activation"},"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":"2510.12229","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-10-14T07:31:29Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"9c4c84e4fb60283a457d2c05f334fce34c8b8331187293abb9c539dc2ee745d4","abstract_canon_sha256":"c205363d632fdb8bf7e1499fa35c4cd850ba733c9d5de980e05398f842643371"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-20T01:18:21.150360Z","signature_b64":"JZCKf+oqq21dyFT4nilVsd07TTp821Ao5yqjZDMGYoYEb66ErEQgv7fUwHAinW8x8N4B7jxKeSuJkH4ye1xWCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"2515293f7b3335a1f2861e1c5ece2d908abae8f02b3fd4a4397d1a0a8b0ba9aa","last_reissued_at":"2026-07-20T01:18:21.149428Z","signature_status":"signed_v1","first_computed_at":"2026-07-20T01:18:21.149428Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Analysing Moral Bias in Finetuned LLMs through Mechanistic Interpretability","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Bianca Raimondi, Daniela Dalbagno, Maurizio Gabbrielli","submitted_at":"2025-10-14T07:31:29Z","abstract_excerpt":"Large language models (LLMs) have been shown to internalize human-like biases during finetuning, yet the mechanisms by which these biases manifest remain unclear. In this work, we investigated whether the well-known Knobe effect, a moral bias in intentionality judgements, emerges in finetuned LLMs and whether it can be traced back to specific components of the model. We conducted a Layer-Patching analysis across 3 open-weights LLMs and demonstrated that the bias is not only learned during finetuning but also localized in a specific set of layers. Surprisingly, we found that patching activation"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2510.12229","kind":"arxiv","version":3},"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/2510.12229/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":"2510.12229","created_at":"2026-07-20T01:18:21.149862+00:00"},{"alias_kind":"arxiv_version","alias_value":"2510.12229v3","created_at":"2026-07-20T01:18:21.149862+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2510.12229","created_at":"2026-07-20T01:18:21.149862+00:00"},{"alias_kind":"pith_short_12","alias_value":"EUKSSP33GM22","created_at":"2026-07-20T01:18:21.149862+00:00"},{"alias_kind":"pith_short_16","alias_value":"EUKSSP33GM22D4UG","created_at":"2026-07-20T01:18:21.149862+00:00"},{"alias_kind":"pith_short_8","alias_value":"EUKSSP33","created_at":"2026-07-20T01:18:21.149862+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.07605","citing_title":"User identity conditions moral wrongness ratings in non-reasoning large language models","ref_index":21,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/EUKSSP33GM22D4UGDYOF5TRNSC","json":"https://pith.science/pith/EUKSSP33GM22D4UGDYOF5TRNSC.json","graph_json":"https://pith.science/api/pith-number/EUKSSP33GM22D4UGDYOF5TRNSC/graph.json","events_json":"https://pith.science/api/pith-number/EUKSSP33GM22D4UGDYOF5TRNSC/events.json","paper":"https://pith.science/paper/EUKSSP33"},"agent_actions":{"view_html":"https://pith.science/pith/EUKSSP33GM22D4UGDYOF5TRNSC","download_json":"https://pith.science/pith/EUKSSP33GM22D4UGDYOF5TRNSC.json","view_paper":"https://pith.science/paper/EUKSSP33","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2510.12229&json=true","fetch_graph":"https://pith.science/api/pith-number/EUKSSP33GM22D4UGDYOF5TRNSC/graph.json","fetch_events":"https://pith.science/api/pith-number/EUKSSP33GM22D4UGDYOF5TRNSC/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/EUKSSP33GM22D4UGDYOF5TRNSC/action/timestamp_anchor","attest_storage":"https://pith.science/pith/EUKSSP33GM22D4UGDYOF5TRNSC/action/storage_attestation","attest_author":"https://pith.science/pith/EUKSSP33GM22D4UGDYOF5TRNSC/action/author_attestation","sign_citation":"https://pith.science/pith/EUKSSP33GM22D4UGDYOF5TRNSC/action/citation_signature","submit_replication":"https://pith.science/pith/EUKSSP33GM22D4UGDYOF5TRNSC/action/replication_record"}},"created_at":"2026-07-20T01:18:21.149862+00:00","updated_at":"2026-07-20T01:18:21.149862+00:00"}