{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2024:DDW74MLXUF6HLVBEYDMIJN4F2T","short_pith_number":"pith:DDW74MLX","canonical_record":{"source":{"id":"2409.12914","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-09-19T17:10:34Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"59e1f4f1589a26832c83127c978b89eb29f27a6341d5d627dc24d982a6b36c61","abstract_canon_sha256":"b175f2134c94be8bdf271119fc1455d6bf85f0e654d962d3637cedff14103e15"},"schema_version":"1.0"},"canonical_sha256":"18edfe3177a17c75d424c0d884b785d4e756b3867baf7fe7601b027620fee71e","source":{"kind":"arxiv","id":"2409.12914","version":3},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2409.12914","created_at":"2026-07-05T10:19:55Z"},{"alias_kind":"arxiv_version","alias_value":"2409.12914v3","created_at":"2026-07-05T10:19:55Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2409.12914","created_at":"2026-07-05T10:19:55Z"},{"alias_kind":"pith_short_12","alias_value":"DDW74MLXUF6H","created_at":"2026-07-05T10:19:55Z"},{"alias_kind":"pith_short_16","alias_value":"DDW74MLXUF6HLVBE","created_at":"2026-07-05T10:19:55Z"},{"alias_kind":"pith_short_8","alias_value":"DDW74MLX","created_at":"2026-07-05T10:19:55Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2024:DDW74MLXUF6HLVBEYDMIJN4F2T","target":"record","payload":{"canonical_record":{"source":{"id":"2409.12914","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-09-19T17:10:34Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"59e1f4f1589a26832c83127c978b89eb29f27a6341d5d627dc24d982a6b36c61","abstract_canon_sha256":"b175f2134c94be8bdf271119fc1455d6bf85f0e654d962d3637cedff14103e15"},"schema_version":"1.0"},"canonical_sha256":"18edfe3177a17c75d424c0d884b785d4e756b3867baf7fe7601b027620fee71e","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:19:55.894078Z","signature_b64":"1zIdASTy0eHkZak62R6l5jnSs/aOPTwrpWrCiHwz5QonQKamnmZgEv/3sBwL5Fxf6+WVZ7WPPEp1jIW47rd/Dg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"18edfe3177a17c75d424c0d884b785d4e756b3867baf7fe7601b027620fee71e","last_reissued_at":"2026-07-05T10:19:55.893603Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:19:55.893603Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2409.12914","source_version":3,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-05T10:19:55Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"hWMSf4Ehck3XyRMfK0gb/6QsKsD9696V7Yn/ssmpOuALJBfeHc3bbEI1kh0DB5Wk2jWgEbdz1qwJ797H5kmzAg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-18T06:56:57.082012Z"},"content_sha256":"ae4999a9debeb5c7dba6aef77abaa458483a6d48e0fffc3b8d0ebb90b51b6aa7","schema_version":"1.0","event_id":"sha256:ae4999a9debeb5c7dba6aef77abaa458483a6d48e0fffc3b8d0ebb90b51b6aa7"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2024:DDW74MLXUF6HLVBEYDMIJN4F2T","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Evaluating Defences against Unsafe Feedback in RLHF","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"David Atanasov, Domenic Rosati, Frank Rudzicz, Giles Edkins, Harsh Raj, Hassan Sajjad, Janarthanan Rajendran, Subhabrata Majumdar","submitted_at":"2024-09-19T17:10:34Z","abstract_excerpt":"While there has been progress towards aligning Large Language Models (LLMs) with human values and ensuring safe behaviour at inference time, safety guards can easily be removed when fine tuned on unsafe and harmful datasets. While this setting has been treated extensively, another popular training paradigm, learning from unsafe feedback with reinforcement learning, has previously been unexplored. This is concerning due to the widespread deployment of feedback collection systems. We address this gap by providing an analysis of learning settings where feedback is harmful, i.e. that unsafe sample"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2409.12914","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/2409.12914/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"},"verdict_id":null},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-05T10:19:55Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"Fq7H3Pk39nYfHa8EWjU1HqOJ0/5FFeaiJD+EU3pBqM0zWhBAVkHIDnhtCFYhAYkdeiiW861+2F5cK4UvC1orBg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-18T06:56:57.082412Z"},"content_sha256":"00017e21edf7cd6ea200d99893c2b9ec666f2aa310f7be350d018446d435dab0","schema_version":"1.0","event_id":"sha256:00017e21edf7cd6ea200d99893c2b9ec666f2aa310f7be350d018446d435dab0"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/DDW74MLXUF6HLVBEYDMIJN4F2T/bundle.json","state_url":"https://pith.science/pith/DDW74MLXUF6HLVBEYDMIJN4F2T/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/DDW74MLXUF6HLVBEYDMIJN4F2T/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":"2026-08-18T06:56:57Z","links":{"resolver":"https://pith.science/pith/DDW74MLXUF6HLVBEYDMIJN4F2T","bundle":"https://pith.science/pith/DDW74MLXUF6HLVBEYDMIJN4F2T/bundle.json","state":"https://pith.science/pith/DDW74MLXUF6HLVBEYDMIJN4F2T/state.json","well_known_bundle":"https://pith.science/.well-known/pith/DDW74MLXUF6HLVBEYDMIJN4F2T/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2024:DDW74MLXUF6HLVBEYDMIJN4F2T","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":"b175f2134c94be8bdf271119fc1455d6bf85f0e654d962d3637cedff14103e15","cross_cats_sorted":["cs.CL"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-09-19T17:10:34Z","title_canon_sha256":"59e1f4f1589a26832c83127c978b89eb29f27a6341d5d627dc24d982a6b36c61"},"schema_version":"1.0","source":{"id":"2409.12914","kind":"arxiv","version":3}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2409.12914","created_at":"2026-07-05T10:19:55Z"},{"alias_kind":"arxiv_version","alias_value":"2409.12914v3","created_at":"2026-07-05T10:19:55Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2409.12914","created_at":"2026-07-05T10:19:55Z"},{"alias_kind":"pith_short_12","alias_value":"DDW74MLXUF6H","created_at":"2026-07-05T10:19:55Z"},{"alias_kind":"pith_short_16","alias_value":"DDW74MLXUF6HLVBE","created_at":"2026-07-05T10:19:55Z"},{"alias_kind":"pith_short_8","alias_value":"DDW74MLX","created_at":"2026-07-05T10:19:55Z"}],"graph_snapshots":[{"event_id":"sha256:00017e21edf7cd6ea200d99893c2b9ec666f2aa310f7be350d018446d435dab0","target":"graph","created_at":"2026-07-05T10:19:55Z","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":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2409.12914/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"While there has been progress towards aligning Large Language Models (LLMs) with human values and ensuring safe behaviour at inference time, safety guards can easily be removed when fine tuned on unsafe and harmful datasets. While this setting has been treated extensively, another popular training paradigm, learning from unsafe feedback with reinforcement learning, has previously been unexplored. This is concerning due to the widespread deployment of feedback collection systems. We address this gap by providing an analysis of learning settings where feedback is harmful, i.e. that unsafe sample","authors_text":"David Atanasov, Domenic Rosati, Frank Rudzicz, Giles Edkins, Harsh Raj, Hassan Sajjad, Janarthanan Rajendran, Subhabrata Majumdar","cross_cats":["cs.CL"],"headline":"","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-09-19T17:10:34Z","title":"Evaluating Defences against Unsafe Feedback in RLHF"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2409.12914","kind":"arxiv","version":3},"verdict":{"created_at":null,"id":null,"model_set":{},"one_line_summary":"","pipeline_version":null,"pith_extraction_headline":"","strongest_claim":"","weakest_assumption":""}},"verdict_id":null}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:ae4999a9debeb5c7dba6aef77abaa458483a6d48e0fffc3b8d0ebb90b51b6aa7","target":"record","created_at":"2026-07-05T10:19:55Z","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":"b175f2134c94be8bdf271119fc1455d6bf85f0e654d962d3637cedff14103e15","cross_cats_sorted":["cs.CL"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-09-19T17:10:34Z","title_canon_sha256":"59e1f4f1589a26832c83127c978b89eb29f27a6341d5d627dc24d982a6b36c61"},"schema_version":"1.0","source":{"id":"2409.12914","kind":"arxiv","version":3}},"canonical_sha256":"18edfe3177a17c75d424c0d884b785d4e756b3867baf7fe7601b027620fee71e","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"18edfe3177a17c75d424c0d884b785d4e756b3867baf7fe7601b027620fee71e","first_computed_at":"2026-07-05T10:19:55.893603Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-05T10:19:55.893603Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"1zIdASTy0eHkZak62R6l5jnSs/aOPTwrpWrCiHwz5QonQKamnmZgEv/3sBwL5Fxf6+WVZ7WPPEp1jIW47rd/Dg==","signature_status":"signed_v1","signed_at":"2026-07-05T10:19:55.894078Z","signed_message":"canonical_sha256_bytes"},"source_id":"2409.12914","source_kind":"arxiv","source_version":3}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:ae4999a9debeb5c7dba6aef77abaa458483a6d48e0fffc3b8d0ebb90b51b6aa7","sha256:00017e21edf7cd6ea200d99893c2b9ec666f2aa310f7be350d018446d435dab0"],"state_sha256":"0d77c3c9a6726a825b8a303f6943c90b726e3977d5779f9aec15f9fd54d24c9c"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"brtczpXwIqFCg0koy6JZDysqOGX42SptXVUTg/e5hjskIWoF7zLTOxykNKHrBKML/Bh8mKU4JjeVrOdpeJdACQ==","signed_message":"bundle_sha256_bytes","signed_at":"2026-08-18T06:56:57.086029Z","bundle_sha256":"7e7659b3ee6d5cc79fed5e158d113602daa34d27918f617f4014bd1c15aa3078"}}