{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2025:ITBVVWVFIP5GNKIAOVFBC3MZNM","short_pith_number":"pith:ITBVVWVF","canonical_record":{"source":{"id":"2502.12206","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2025-02-16T16:29:20Z","cross_cats_sorted":["cs.CL","cs.LG"],"title_canon_sha256":"57e46e5977200487f549729b0c81a2e0fcc23893fde017255d31858d91dbbf45","abstract_canon_sha256":"8bbfb903970437f5ff9de515f93460c83b01d84f9aa9fd2d79181eecb0f088ee"},"schema_version":"1.0"},"canonical_sha256":"44c35adaa543fa66a900754a116d996b0683a4f2c5decf49dd8756814aaed401","source":{"kind":"arxiv","id":"2502.12206","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2502.12206","created_at":"2026-07-05T10:15:51Z"},{"alias_kind":"arxiv_version","alias_value":"2502.12206v1","created_at":"2026-07-05T10:15:51Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.12206","created_at":"2026-07-05T10:15:51Z"},{"alias_kind":"pith_short_12","alias_value":"ITBVVWVFIP5G","created_at":"2026-07-05T10:15:51Z"},{"alias_kind":"pith_short_16","alias_value":"ITBVVWVFIP5GNKIA","created_at":"2026-07-05T10:15:51Z"},{"alias_kind":"pith_short_8","alias_value":"ITBVVWVF","created_at":"2026-07-05T10:15:51Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2025:ITBVVWVFIP5GNKIAOVFBC3MZNM","target":"record","payload":{"canonical_record":{"source":{"id":"2502.12206","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2025-02-16T16:29:20Z","cross_cats_sorted":["cs.CL","cs.LG"],"title_canon_sha256":"57e46e5977200487f549729b0c81a2e0fcc23893fde017255d31858d91dbbf45","abstract_canon_sha256":"8bbfb903970437f5ff9de515f93460c83b01d84f9aa9fd2d79181eecb0f088ee"},"schema_version":"1.0"},"canonical_sha256":"44c35adaa543fa66a900754a116d996b0683a4f2c5decf49dd8756814aaed401","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:15:51.917650Z","signature_b64":"JC3zRecpKKLTTl03SpcKckkBRxyU0lHB/H1KNQ08VUZ7gPNYoiL4FAa6hFVlVEk/zXXXD94HLO9jpYt6J1/LDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"44c35adaa543fa66a900754a116d996b0683a4f2c5decf49dd8756814aaed401","last_reissued_at":"2026-07-05T10:15:51.917123Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:15:51.917123Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2502.12206","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":"2026-07-05T10:15:51Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"JR7DdTzAVjOtQLP7qaqV35hi8qm35NMBmBfAUqOMwh06QD2PYHDNTHlZnp2ki5kDvonlV/i8+rDkGM4IdkwLBA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-08T16:11:31.909202Z"},"content_sha256":"2cf731cc039b916d6729356eb648ddc1f9499f42e844fb75259ef5ba1b4ed2c5","schema_version":"1.0","event_id":"sha256:2cf731cc039b916d6729356eb648ddc1f9499f42e844fb75259ef5ba1b4ed2c5"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2025:ITBVVWVFIP5GNKIAOVFBC3MZNM","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Evaluating the Paperclip Maximizer: Are RL-Based Language Models More Likely to Pursue Instrumental Goals?","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL","cs.LG"],"primary_cat":"cs.AI","authors_text":"Bryan Hooi, Jiaying Wu, Yuan Sui, Yuexin Li, Yufei He, Yulin Chen","submitted_at":"2025-02-16T16:29:20Z","abstract_excerpt":"As large language models (LLMs) continue to evolve, ensuring their alignment with human goals and values remains a pressing challenge. A key concern is \\textit{instrumental convergence}, where an AI system, in optimizing for a given objective, develops unintended intermediate goals that override the ultimate objective and deviate from human-intended goals. This issue is particularly relevant in reinforcement learning (RL)-trained models, which can generate creative but unintended strategies to maximize rewards. In this paper, we explore instrumental convergence in LLMs by comparing models trai"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.12206","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/2502.12206/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":"2026-07-05T10:15:51Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"2YjYquwtR4Hdb25pg09aNI2NorbVH+q7N+B8nCB3MXT/TmFZPbNrr3Yha3sCgBNuztVCiPfIfwbBf7MwBjhGDg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-08T16:11:31.909781Z"},"content_sha256":"cbc5edefe6018cdb92f29506cdd3e9f4b20fb5edd34d93eef1d2c56490faf36d","schema_version":"1.0","event_id":"sha256:cbc5edefe6018cdb92f29506cdd3e9f4b20fb5edd34d93eef1d2c56490faf36d"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/ITBVVWVFIP5GNKIAOVFBC3MZNM/bundle.json","state_url":"https://pith.science/pith/ITBVVWVFIP5GNKIAOVFBC3MZNM/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/ITBVVWVFIP5GNKIAOVFBC3MZNM/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":"2026-08-08T16:11:31Z","links":{"resolver":"https://pith.science/pith/ITBVVWVFIP5GNKIAOVFBC3MZNM","bundle":"https://pith.science/pith/ITBVVWVFIP5GNKIAOVFBC3MZNM/bundle.json","state":"https://pith.science/pith/ITBVVWVFIP5GNKIAOVFBC3MZNM/state.json","well_known_bundle":"https://pith.science/.well-known/pith/ITBVVWVFIP5GNKIAOVFBC3MZNM/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2025:ITBVVWVFIP5GNKIAOVFBC3MZNM","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":"8bbfb903970437f5ff9de515f93460c83b01d84f9aa9fd2d79181eecb0f088ee","cross_cats_sorted":["cs.CL","cs.LG"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2025-02-16T16:29:20Z","title_canon_sha256":"57e46e5977200487f549729b0c81a2e0fcc23893fde017255d31858d91dbbf45"},"schema_version":"1.0","source":{"id":"2502.12206","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2502.12206","created_at":"2026-07-05T10:15:51Z"},{"alias_kind":"arxiv_version","alias_value":"2502.12206v1","created_at":"2026-07-05T10:15:51Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.12206","created_at":"2026-07-05T10:15:51Z"},{"alias_kind":"pith_short_12","alias_value":"ITBVVWVFIP5G","created_at":"2026-07-05T10:15:51Z"},{"alias_kind":"pith_short_16","alias_value":"ITBVVWVFIP5GNKIA","created_at":"2026-07-05T10:15:51Z"},{"alias_kind":"pith_short_8","alias_value":"ITBVVWVF","created_at":"2026-07-05T10:15:51Z"}],"graph_snapshots":[{"event_id":"sha256:cbc5edefe6018cdb92f29506cdd3e9f4b20fb5edd34d93eef1d2c56490faf36d","target":"graph","created_at":"2026-07-05T10:15:51Z","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/2502.12206/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"As large language models (LLMs) continue to evolve, ensuring their alignment with human goals and values remains a pressing challenge. A key concern is \\textit{instrumental convergence}, where an AI system, in optimizing for a given objective, develops unintended intermediate goals that override the ultimate objective and deviate from human-intended goals. This issue is particularly relevant in reinforcement learning (RL)-trained models, which can generate creative but unintended strategies to maximize rewards. In this paper, we explore instrumental convergence in LLMs by comparing models trai","authors_text":"Bryan Hooi, Jiaying Wu, Yuan Sui, Yuexin Li, Yufei He, Yulin Chen","cross_cats":["cs.CL","cs.LG"],"headline":"","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2025-02-16T16:29:20Z","title":"Evaluating the Paperclip Maximizer: Are RL-Based Language Models More Likely to Pursue Instrumental Goals?"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.12206","kind":"arxiv","version":1},"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":"sha256:2cf731cc039b916d6729356eb648ddc1f9499f42e844fb75259ef5ba1b4ed2c5","target":"record","created_at":"2026-07-05T10:15:51Z","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":"8bbfb903970437f5ff9de515f93460c83b01d84f9aa9fd2d79181eecb0f088ee","cross_cats_sorted":["cs.CL","cs.LG"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2025-02-16T16:29:20Z","title_canon_sha256":"57e46e5977200487f549729b0c81a2e0fcc23893fde017255d31858d91dbbf45"},"schema_version":"1.0","source":{"id":"2502.12206","kind":"arxiv","version":1}},"canonical_sha256":"44c35adaa543fa66a900754a116d996b0683a4f2c5decf49dd8756814aaed401","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"44c35adaa543fa66a900754a116d996b0683a4f2c5decf49dd8756814aaed401","first_computed_at":"2026-07-05T10:15:51.917123Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-05T10:15:51.917123Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"JC3zRecpKKLTTl03SpcKckkBRxyU0lHB/H1KNQ08VUZ7gPNYoiL4FAa6hFVlVEk/zXXXD94HLO9jpYt6J1/LDQ==","signature_status":"signed_v1","signed_at":"2026-07-05T10:15:51.917650Z","signed_message":"canonical_sha256_bytes"},"source_id":"2502.12206","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:2cf731cc039b916d6729356eb648ddc1f9499f42e844fb75259ef5ba1b4ed2c5","sha256:cbc5edefe6018cdb92f29506cdd3e9f4b20fb5edd34d93eef1d2c56490faf36d"],"state_sha256":"3b21047674c3c09362a9a2168722b6edf95ecfae0da966da77fad9d536bdf9b1"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"GffMH1gMsMnQGkKcgzdrevt2l5nmv+LR7RQ/NObcWPo0MZyCNxz0vyhtitDQ1zIn1xIxmDi4nxlJo3+yw6uTBQ==","signed_message":"bundle_sha256_bytes","signed_at":"2026-08-08T16:11:31.914161Z","bundle_sha256":"a8abe1338c3491927158d54fdf8c0364e185860bf1aa8f9b6fd70abbb01cf980"}}