{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:7FE2GNCIG7CKXWKHRH7PHZZFDD","short_pith_number":"pith:7FE2GNCI","schema_version":"1.0","canonical_sha256":"f949a3344837c4abd94789fef3e72518ea993202f13b6f138400b8c46ddc7329","source":{"kind":"arxiv","id":"2505.23561","version":1},"attestation_state":"computed","paper":{"title":"Merge Hijacking: Backdoor Attacks to Model Merging of Large Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CR","authors_text":"Jiawen Shi, Lichao Sun, Pan Zhou, Yangming Xu, Zenghui Yuan","submitted_at":"2025-05-29T15:37:23Z","abstract_excerpt":"Model merging for Large Language Models (LLMs) directly fuses the parameters of different models finetuned on various tasks, creating a unified model for multi-domain tasks. However, due to potential vulnerabilities in models available on open-source platforms, model merging is susceptible to backdoor attacks. In this paper, we propose Merge Hijacking, the first backdoor attack targeting model merging in LLMs. The attacker constructs a malicious upload model and releases it. Once a victim user merges it with any other models, the resulting merged model inherits the backdoor while maintaining u"},"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":"2505.23561","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CR","submitted_at":"2025-05-29T15:37:23Z","cross_cats_sorted":[],"title_canon_sha256":"fe871070d517af4049526203f768dd886d2f49abd41be2030f6c5824fd3f499e","abstract_canon_sha256":"0a9caf4dbd9efa530423ea8d42e1ad01d870fdc82dbdb114c78f2f929fd4bf46"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:12:03.969382Z","signature_b64":"A5a4LWg0DBU5sVgUr4hdMdcyZ+qzXpFIAi/6pVCRNgVVbgjPhATTc09wjQ5Uo07OvBQ3EBpdn3zE4zB4Hm4zDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"f949a3344837c4abd94789fef3e72518ea993202f13b6f138400b8c46ddc7329","last_reissued_at":"2026-07-05T11:12:03.968877Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:12:03.968877Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Merge Hijacking: Backdoor Attacks to Model Merging of Large Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CR","authors_text":"Jiawen Shi, Lichao Sun, Pan Zhou, Yangming Xu, Zenghui Yuan","submitted_at":"2025-05-29T15:37:23Z","abstract_excerpt":"Model merging for Large Language Models (LLMs) directly fuses the parameters of different models finetuned on various tasks, creating a unified model for multi-domain tasks. However, due to potential vulnerabilities in models available on open-source platforms, model merging is susceptible to backdoor attacks. In this paper, we propose Merge Hijacking, the first backdoor attack targeting model merging in LLMs. The attacker constructs a malicious upload model and releases it. Once a victim user merges it with any other models, the resulting merged model inherits the backdoor while maintaining u"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.23561","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/2505.23561/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":"2505.23561","created_at":"2026-07-05T11:12:03.968935+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.23561v1","created_at":"2026-07-05T11:12:03.968935+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.23561","created_at":"2026-07-05T11:12:03.968935+00:00"},{"alias_kind":"pith_short_12","alias_value":"7FE2GNCIG7CK","created_at":"2026-07-05T11:12:03.968935+00:00"},{"alias_kind":"pith_short_16","alias_value":"7FE2GNCIG7CKXWKH","created_at":"2026-07-05T11:12:03.968935+00:00"},{"alias_kind":"pith_short_8","alias_value":"7FE2GNCI","created_at":"2026-07-05T11:12:03.968935+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.12498","citing_title":"From Parameters to Feature Space: Task Arithmetic for Backdoor Mitigation in Model Merging","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2606.03344","citing_title":"RogueMerge: Robust and Unified Attacks against LLM Model Merging","ref_index":14,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/7FE2GNCIG7CKXWKHRH7PHZZFDD","json":"https://pith.science/pith/7FE2GNCIG7CKXWKHRH7PHZZFDD.json","graph_json":"https://pith.science/api/pith-number/7FE2GNCIG7CKXWKHRH7PHZZFDD/graph.json","events_json":"https://pith.science/api/pith-number/7FE2GNCIG7CKXWKHRH7PHZZFDD/events.json","paper":"https://pith.science/paper/7FE2GNCI"},"agent_actions":{"view_html":"https://pith.science/pith/7FE2GNCIG7CKXWKHRH7PHZZFDD","download_json":"https://pith.science/pith/7FE2GNCIG7CKXWKHRH7PHZZFDD.json","view_paper":"https://pith.science/paper/7FE2GNCI","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.23561&json=true","fetch_graph":"https://pith.science/api/pith-number/7FE2GNCIG7CKXWKHRH7PHZZFDD/graph.json","fetch_events":"https://pith.science/api/pith-number/7FE2GNCIG7CKXWKHRH7PHZZFDD/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/7FE2GNCIG7CKXWKHRH7PHZZFDD/action/timestamp_anchor","attest_storage":"https://pith.science/pith/7FE2GNCIG7CKXWKHRH7PHZZFDD/action/storage_attestation","attest_author":"https://pith.science/pith/7FE2GNCIG7CKXWKHRH7PHZZFDD/action/author_attestation","sign_citation":"https://pith.science/pith/7FE2GNCIG7CKXWKHRH7PHZZFDD/action/citation_signature","submit_replication":"https://pith.science/pith/7FE2GNCIG7CKXWKHRH7PHZZFDD/action/replication_record"}},"created_at":"2026-07-05T11:12:03.968935+00:00","updated_at":"2026-07-05T11:12:03.968935+00:00"}