{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:V5OHABJBMTHZ2LU4YCMN3YXRGC","short_pith_number":"pith:V5OHABJB","canonical_record":{"source":{"id":"2605.14514","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CR","submitted_at":"2026-05-14T07:58:47Z","cross_cats_sorted":[],"title_canon_sha256":"6a874cabc0ac37f88231d9ef0228f568917a81b0d931f95e85919befbe2d9a69","abstract_canon_sha256":"14656f567a83f12c672bb60f699e005b12149747998e20e2304550abf590edd9"},"schema_version":"1.0"},"canonical_sha256":"af5c70052164cf9d2e9cc098dde2f13094583a4dc8a6b9b317c5583b7ce3673b","source":{"kind":"arxiv","id":"2605.14514","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.14514","created_at":"2026-05-17T23:39:06Z"},{"alias_kind":"arxiv_version","alias_value":"2605.14514v1","created_at":"2026-05-17T23:39:06Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.14514","created_at":"2026-05-17T23:39:06Z"},{"alias_kind":"pith_short_12","alias_value":"V5OHABJBMTHZ","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"V5OHABJBMTHZ2LU4","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"V5OHABJB","created_at":"2026-05-18T12:33:37Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:V5OHABJBMTHZ2LU4YCMN3YXRGC","target":"record","payload":{"canonical_record":{"source":{"id":"2605.14514","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CR","submitted_at":"2026-05-14T07:58:47Z","cross_cats_sorted":[],"title_canon_sha256":"6a874cabc0ac37f88231d9ef0228f568917a81b0d931f95e85919befbe2d9a69","abstract_canon_sha256":"14656f567a83f12c672bb60f699e005b12149747998e20e2304550abf590edd9"},"schema_version":"1.0"},"canonical_sha256":"af5c70052164cf9d2e9cc098dde2f13094583a4dc8a6b9b317c5583b7ce3673b","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-17T23:39:06.155241Z","signature_b64":"Je8JF3KOxaDI6j3pU6MwAzNoxvO1B3VZ7a1CJIG7DVCTk4eokQ4UmHUzl+89uPIBXzGWVgJrkiGQqD0o4YgwAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"af5c70052164cf9d2e9cc098dde2f13094583a4dc8a6b9b317c5583b7ce3673b","last_reissued_at":"2026-05-17T23:39:06.154548Z","signature_status":"signed_v1","first_computed_at":"2026-05-17T23:39:06.154548Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2605.14514","source_version":1,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-17T23:39:06Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"Q+4tVxyKvx1LfOyENlZp20Rt1OD4LI0Z4GJyC0hzPtNNr+OTkTdHifwcXToLiuF4Ldo9ueTU93OOsdvpK126AQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-26T05:09:46.559272Z"},"content_sha256":"a2e4833479c671689ffee84ff6cde9d145e80548bd0a2479e63f4e955d2a02e9","schema_version":"1.0","event_id":"sha256:a2e4833479c671689ffee84ff6cde9d145e80548bd0a2479e63f4e955d2a02e9"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:V5OHABJBMTHZ2LU4YCMN3YXRGC","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Defenses at Odds: Measuring and Explaining Defense Conflicts in Large Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Sequential safety defenses on large language models often undermine earlier protections in 38.9 percent of deployment orders.","cross_cats":[],"primary_cat":"cs.CR","authors_text":"Chuanchao Zang, Jianing Wang, Li Wang, Shanqing Guo, Wenyu Chen, Xiangtao Meng, Xinyu Gao, Zheng Li","submitted_at":"2026-05-14T07:58:47Z","abstract_excerpt":"Large Language Models (LLMs) deployed in high-stakes applications must simultaneously manage multiple risks, yet existing defenses are almost exclusively evaluated in isolation under a one-shot deployment assumption. In practice, providers patch models incrementally throughout their lifecycle-responding to newly exposed vulnerabilities or targeted data-removal requests without retraining from scratch. This raises a fundamental but underexplored question: does a later defense preserve the protections established by an earlier one? We present the first systematic study of cross-defense interacti"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"38.9% of 144 ordered sequences exhibit measurable risk exacerbation on the originally defended dimension. These interactions are highly asymmetric and order-dependent.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the chosen risk dimensions, evaluation metrics, and sequential patching without retraining accurately model real-world multi-defense deployment and capture true safety properties.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Sequential LLM defense deployment leads to risk exacerbation in 38.9% of cases due to anti-aligned updates in shared critical layers, addressed by conflict-guided layer freezing.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Sequential safety defenses on large language models often undermine earlier protections in 38.9 percent of deployment orders.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"a7a3af3b1d7d58a18f8bd16bbfb26ba20e1ecfb4c88e1680b35e6cbdf8751386"},"source":{"id":"2605.14514","kind":"arxiv","version":1},"verdict":{"id":"17ffc144-d345-4aa4-976a-6722376c5f90","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T01:39:24.380910Z","strongest_claim":"38.9% of 144 ordered sequences exhibit measurable risk exacerbation on the originally defended dimension. These interactions are highly asymmetric and order-dependent.","one_line_summary":"Sequential LLM defense deployment leads to risk exacerbation in 38.9% of cases due to anti-aligned updates in shared critical layers, addressed by conflict-guided layer freezing.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the chosen risk dimensions, evaluation metrics, and sequential patching without retraining accurately model real-world multi-defense deployment and capture true safety properties.","pith_extraction_headline":"Sequential safety defenses on large language models often undermine earlier protections in 38.9 percent of deployment orders."},"references":{"count":57,"sample":[{"doi":"","year":2016,"title":"Deep learning with differential privacy","work_id":"e0402a84-2de7-43d9-bb5e-023ab2d71c6a","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2022,"title":"Constitutional AI: Harmlessness from AI Feedback","work_id":"faaaa4e0-2676-4fac-a0b4-99aef10d2095","ref_index":2,"cited_arxiv_id":"2212.08073","is_internal_anchor":true},{"doi":"","year":2021,"title":"Machine un- learning","work_id":"20cc13d8-81da-4d20-87d3-cf5a8cded160","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2021,"title":"Extracting training data from large language mod- els","work_id":"565de915-35f7-402a-99c9-5d70387132bf","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"Safe RLHF: Safe Reinforcement Learning from Human Feedback","work_id":"45bf2da5-62a0-441e-b0fc-77d659b681db","ref_index":5,"cited_arxiv_id":"2310.12773","is_internal_anchor":true}],"resolved_work":57,"snapshot_sha256":"2a83c39ca8ddc7ef14e1367736e1ba5dcf6eefe219ce15c2e9b49d4605ed21fd","internal_anchors":14},"formal_canon":{"evidence_count":2,"snapshot_sha256":"2f61a8cb3b09db9edb867b3e85e9f39aad4290ed955a13460ffebf793581a872"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"verdict_id":"17ffc144-d345-4aa4-976a-6722376c5f90"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-17T23:39:06Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"3WeSpjhtNZroEedDr8NROq5n1Ws5PxXK/Oo42v7R13j083HR7tbSZVKbNwdSR1iokx/+Wp8m4riOXIg/uCCQCQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-26T05:09:46.559921Z"},"content_sha256":"c82759b7cc650f50f0e3ed049fef35a4b6fa335042eb34f6ff0f066d19674c60","schema_version":"1.0","event_id":"sha256:c82759b7cc650f50f0e3ed049fef35a4b6fa335042eb34f6ff0f066d19674c60"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/V5OHABJBMTHZ2LU4YCMN3YXRGC/bundle.json","state_url":"https://pith.science/pith/V5OHABJBMTHZ2LU4YCMN3YXRGC/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/V5OHABJBMTHZ2LU4YCMN3YXRGC/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-05-26T05:09:46Z","links":{"resolver":"https://pith.science/pith/V5OHABJBMTHZ2LU4YCMN3YXRGC","bundle":"https://pith.science/pith/V5OHABJBMTHZ2LU4YCMN3YXRGC/bundle.json","state":"https://pith.science/pith/V5OHABJBMTHZ2LU4YCMN3YXRGC/state.json","well_known_bundle":"https://pith.science/.well-known/pith/V5OHABJBMTHZ2LU4YCMN3YXRGC/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:V5OHABJBMTHZ2LU4YCMN3YXRGC","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"14656f567a83f12c672bb60f699e005b12149747998e20e2304550abf590edd9","cross_cats_sorted":[],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CR","submitted_at":"2026-05-14T07:58:47Z","title_canon_sha256":"6a874cabc0ac37f88231d9ef0228f568917a81b0d931f95e85919befbe2d9a69"},"schema_version":"1.0","source":{"id":"2605.14514","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.14514","created_at":"2026-05-17T23:39:06Z"},{"alias_kind":"arxiv_version","alias_value":"2605.14514v1","created_at":"2026-05-17T23:39:06Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.14514","created_at":"2026-05-17T23:39:06Z"},{"alias_kind":"pith_short_12","alias_value":"V5OHABJBMTHZ","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"V5OHABJBMTHZ2LU4","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"V5OHABJB","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:c82759b7cc650f50f0e3ed049fef35a4b6fa335042eb34f6ff0f066d19674c60","target":"graph","created_at":"2026-05-17T23:39:06Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"38.9% of 144 ordered sequences exhibit measurable risk exacerbation on the originally defended dimension. These interactions are highly asymmetric and order-dependent."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the chosen risk dimensions, evaluation metrics, and sequential patching without retraining accurately model real-world multi-defense deployment and capture true safety properties."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"Sequential LLM defense deployment leads to risk exacerbation in 38.9% of cases due to anti-aligned updates in shared critical layers, addressed by conflict-guided layer freezing."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Sequential safety defenses on large language models often undermine earlier protections in 38.9 percent of deployment orders."}],"snapshot_sha256":"a7a3af3b1d7d58a18f8bd16bbfb26ba20e1ecfb4c88e1680b35e6cbdf8751386"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"2f61a8cb3b09db9edb867b3e85e9f39aad4290ed955a13460ffebf793581a872"},"paper":{"abstract_excerpt":"Large Language Models (LLMs) deployed in high-stakes applications must simultaneously manage multiple risks, yet existing defenses are almost exclusively evaluated in isolation under a one-shot deployment assumption. In practice, providers patch models incrementally throughout their lifecycle-responding to newly exposed vulnerabilities or targeted data-removal requests without retraining from scratch. This raises a fundamental but underexplored question: does a later defense preserve the protections established by an earlier one? We present the first systematic study of cross-defense interacti","authors_text":"Chuanchao Zang, Jianing Wang, Li Wang, Shanqing Guo, Wenyu Chen, Xiangtao Meng, Xinyu Gao, Zheng Li","cross_cats":[],"headline":"Sequential safety defenses on large language models often undermine earlier protections in 38.9 percent of deployment orders.","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CR","submitted_at":"2026-05-14T07:58:47Z","title":"Defenses at Odds: Measuring and Explaining Defense Conflicts in Large Language Models"},"references":{"count":57,"internal_anchors":14,"resolved_work":57,"sample":[{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":1,"title":"Deep learning with differential privacy","work_id":"e0402a84-2de7-43d9-bb5e-023ab2d71c6a","year":2016},{"cited_arxiv_id":"2212.08073","doi":"","is_internal_anchor":true,"ref_index":2,"title":"Constitutional AI: Harmlessness from AI Feedback","work_id":"faaaa4e0-2676-4fac-a0b4-99aef10d2095","year":2022},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"Machine un- learning","work_id":"20cc13d8-81da-4d20-87d3-cf5a8cded160","year":2021},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"Extracting training data from large language mod- els","work_id":"565de915-35f7-402a-99c9-5d70387132bf","year":2021},{"cited_arxiv_id":"2310.12773","doi":"","is_internal_anchor":true,"ref_index":5,"title":"Safe RLHF: Safe Reinforcement Learning from Human Feedback","work_id":"45bf2da5-62a0-441e-b0fc-77d659b681db","year":2023}],"snapshot_sha256":"2a83c39ca8ddc7ef14e1367736e1ba5dcf6eefe219ce15c2e9b49d4605ed21fd"},"source":{"id":"2605.14514","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-15T01:39:24.380910Z","id":"17ffc144-d345-4aa4-976a-6722376c5f90","model_set":{"reader":"grok-4.3"},"one_line_summary":"Sequential LLM defense deployment leads to risk exacerbation in 38.9% of cases due to anti-aligned updates in shared critical layers, addressed by conflict-guided layer freezing.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Sequential safety defenses on large language models often undermine earlier protections in 38.9 percent of deployment orders.","strongest_claim":"38.9% of 144 ordered sequences exhibit measurable risk exacerbation on the originally defended dimension. These interactions are highly asymmetric and order-dependent.","weakest_assumption":"That the chosen risk dimensions, evaluation metrics, and sequential patching without retraining accurately model real-world multi-defense deployment and capture true safety properties."}},"verdict_id":"17ffc144-d345-4aa4-976a-6722376c5f90"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:a2e4833479c671689ffee84ff6cde9d145e80548bd0a2479e63f4e955d2a02e9","target":"record","created_at":"2026-05-17T23:39:06Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"14656f567a83f12c672bb60f699e005b12149747998e20e2304550abf590edd9","cross_cats_sorted":[],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CR","submitted_at":"2026-05-14T07:58:47Z","title_canon_sha256":"6a874cabc0ac37f88231d9ef0228f568917a81b0d931f95e85919befbe2d9a69"},"schema_version":"1.0","source":{"id":"2605.14514","kind":"arxiv","version":1}},"canonical_sha256":"af5c70052164cf9d2e9cc098dde2f13094583a4dc8a6b9b317c5583b7ce3673b","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"af5c70052164cf9d2e9cc098dde2f13094583a4dc8a6b9b317c5583b7ce3673b","first_computed_at":"2026-05-17T23:39:06.154548Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-17T23:39:06.154548Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"Je8JF3KOxaDI6j3pU6MwAzNoxvO1B3VZ7a1CJIG7DVCTk4eokQ4UmHUzl+89uPIBXzGWVgJrkiGQqD0o4YgwAw==","signature_status":"signed_v1","signed_at":"2026-05-17T23:39:06.155241Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.14514","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:a2e4833479c671689ffee84ff6cde9d145e80548bd0a2479e63f4e955d2a02e9","sha256:c82759b7cc650f50f0e3ed049fef35a4b6fa335042eb34f6ff0f066d19674c60"],"state_sha256":"bfd0050bdf488f9b00322a2903f83f5aa015ca89278f990c3d94ee1a627a35e5"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"QBjpsI/hMD3Iel4JioFDNK/lo8tGaZYCKfj5F8cjXLYnCiltSnYaYM7ivlcRaKjBmq0Bky2EhqOlJKtQsZqNDw==","signed_message":"bundle_sha256_bytes","signed_at":"2026-05-26T05:09:46.562797Z","bundle_sha256":"11770738f7616a097b87e764b13f0583e939a9def700ce2655d80119f3633ade"}}