{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:C2UELM7WLDE6EBOLVRPJVUY6WI","short_pith_number":"pith:C2UELM7W","schema_version":"1.0","canonical_sha256":"16a845b3f658c9e205cbac5e9ad31eb21103b999be59eb6e4e2890b5eb3a19a3","source":{"kind":"arxiv","id":"2508.09019","version":1},"attestation_state":"computed","paper":{"title":"Activation Steering for Bias Mitigation: An Interpretable Approach to Safer LLMs","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Shivam Dubey","submitted_at":"2025-08-12T15:34:18Z","abstract_excerpt":"As large language models (LLMs) become more integrated into societal systems, the risk of them perpetuating and amplifying harmful biases becomes a critical safety concern. Traditional methods for mitigating bias often rely on data filtering or post-hoc output moderation, which treat the model as an opaque black box. In this work, we introduce a complete, end-to-end system that uses techniques from mechanistic interpretability to both identify and actively mitigate bias directly within a model's internal workings. Our method involves two primary stages. First, we train linear \"probes\" on the i"},"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":"2508.09019","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.AI","submitted_at":"2025-08-12T15:34:18Z","cross_cats_sorted":[],"title_canon_sha256":"7d7c1461b0f96a1c2a8d90ce660167f63efad70faf742f4bcef564d40150c7f1","abstract_canon_sha256":"d979a423e71e34ada28f448bed45c263ee0ba8237cd90f9c13cbb1b7bf797894"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:52:38.461684Z","signature_b64":"I49mgLd6kWTPVsVTc6VZHh17XZ+zy3tvxftS0A5TYRvSXUU03dVDVXEPdHpvbthIK4/Gh6MHJZ3Nt+lbjqD3Cw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"16a845b3f658c9e205cbac5e9ad31eb21103b999be59eb6e4e2890b5eb3a19a3","last_reissued_at":"2026-07-05T11:52:38.461196Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:52:38.461196Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Activation Steering for Bias Mitigation: An Interpretable Approach to Safer LLMs","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Shivam Dubey","submitted_at":"2025-08-12T15:34:18Z","abstract_excerpt":"As large language models (LLMs) become more integrated into societal systems, the risk of them perpetuating and amplifying harmful biases becomes a critical safety concern. Traditional methods for mitigating bias often rely on data filtering or post-hoc output moderation, which treat the model as an opaque black box. In this work, we introduce a complete, end-to-end system that uses techniques from mechanistic interpretability to both identify and actively mitigate bias directly within a model's internal workings. Our method involves two primary stages. First, we train linear \"probes\" on the i"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2508.09019","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/2508.09019/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":"2508.09019","created_at":"2026-07-05T11:52:38.461255+00:00"},{"alias_kind":"arxiv_version","alias_value":"2508.09019v1","created_at":"2026-07-05T11:52:38.461255+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2508.09019","created_at":"2026-07-05T11:52:38.461255+00:00"},{"alias_kind":"pith_short_12","alias_value":"C2UELM7WLDE6","created_at":"2026-07-05T11:52:38.461255+00:00"},{"alias_kind":"pith_short_16","alias_value":"C2UELM7WLDE6EBOL","created_at":"2026-07-05T11:52:38.461255+00:00"},{"alias_kind":"pith_short_8","alias_value":"C2UELM7W","created_at":"2026-07-05T11:52:38.461255+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/C2UELM7WLDE6EBOLVRPJVUY6WI","json":"https://pith.science/pith/C2UELM7WLDE6EBOLVRPJVUY6WI.json","graph_json":"https://pith.science/api/pith-number/C2UELM7WLDE6EBOLVRPJVUY6WI/graph.json","events_json":"https://pith.science/api/pith-number/C2UELM7WLDE6EBOLVRPJVUY6WI/events.json","paper":"https://pith.science/paper/C2UELM7W"},"agent_actions":{"view_html":"https://pith.science/pith/C2UELM7WLDE6EBOLVRPJVUY6WI","download_json":"https://pith.science/pith/C2UELM7WLDE6EBOLVRPJVUY6WI.json","view_paper":"https://pith.science/paper/C2UELM7W","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2508.09019&json=true","fetch_graph":"https://pith.science/api/pith-number/C2UELM7WLDE6EBOLVRPJVUY6WI/graph.json","fetch_events":"https://pith.science/api/pith-number/C2UELM7WLDE6EBOLVRPJVUY6WI/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/C2UELM7WLDE6EBOLVRPJVUY6WI/action/timestamp_anchor","attest_storage":"https://pith.science/pith/C2UELM7WLDE6EBOLVRPJVUY6WI/action/storage_attestation","attest_author":"https://pith.science/pith/C2UELM7WLDE6EBOLVRPJVUY6WI/action/author_attestation","sign_citation":"https://pith.science/pith/C2UELM7WLDE6EBOLVRPJVUY6WI/action/citation_signature","submit_replication":"https://pith.science/pith/C2UELM7WLDE6EBOLVRPJVUY6WI/action/replication_record"}},"created_at":"2026-07-05T11:52:38.461255+00:00","updated_at":"2026-07-05T11:52:38.461255+00:00"}