{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:5BPXGGNB4ZRGLACOBJVMZENMMZ","short_pith_number":"pith:5BPXGGNB","schema_version":"1.0","canonical_sha256":"e85f7319a1e66265804e0a6acc91ac6645cda8694c5c98797bc630572d5471db","source":{"kind":"arxiv","id":"2608.00218","version":1},"attestation_state":"computed","paper":{"title":"A Few Neurons Reveal When LLMs Misuse Tools: Sparse Detection and Selective Steering for Reliable Tool Use","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Chongwen Zhao, Kaizhu Huang, Ming Yin, Yutong Ke","submitted_at":"2026-07-31T19:03:44Z","abstract_excerpt":"Agentic LLMs exhibit three consequential tool-use failures: invalid arguments (validity), unnecessary calls (over-calling), and omitted calls when tools are needed (missing). We find that a small, failure-specific set of MLP neurons could distinguish such failures with linearly separable decision boundaries. Building on this observation, we introduce PRISMS (Probing Representations In Support of Monitoring and Steering), a closed-loop framework that shares a failure-specific neuron basis between sparse detection and activation steering. PRISMS selects contribution-critical MLP neurons and fits"},"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":"2608.00218","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2026-07-31T19:03:44Z","cross_cats_sorted":[],"title_canon_sha256":"9985fb05c3adf0ec281c201fc60891132d9b34b83a7513940e562adeb9eab178","abstract_canon_sha256":"8e88736e35b5c1a4f80867486c7d975a40dab9e027907c6f2fa249d4b6452bbe"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-08-04T00:34:25.665336Z","signature_b64":"jkYQ1eK1y9EBKSPrlXJGsIV3PsudEtO6L+4TEz6vFU2iUTgyXnTWmzWzdwNtfoz7wSN4Oj+Wmqou8gkdHtbdBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"e85f7319a1e66265804e0a6acc91ac6645cda8694c5c98797bc630572d5471db","last_reissued_at":"2026-08-04T00:34:25.663880Z","signature_status":"signed_v1","first_computed_at":"2026-08-04T00:34:25.663880Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"A Few Neurons Reveal When LLMs Misuse Tools: Sparse Detection and Selective Steering for Reliable Tool Use","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Chongwen Zhao, Kaizhu Huang, Ming Yin, Yutong Ke","submitted_at":"2026-07-31T19:03:44Z","abstract_excerpt":"Agentic LLMs exhibit three consequential tool-use failures: invalid arguments (validity), unnecessary calls (over-calling), and omitted calls when tools are needed (missing). We find that a small, failure-specific set of MLP neurons could distinguish such failures with linearly separable decision boundaries. Building on this observation, we introduce PRISMS (Probing Representations In Support of Monitoring and Steering), a closed-loop framework that shares a failure-specific neuron basis between sparse detection and activation steering. PRISMS selects contribution-critical MLP neurons and fits"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2608.00218","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/2608.00218/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":"2608.00218","created_at":"2026-08-04T00:34:25.664986+00:00"},{"alias_kind":"arxiv_version","alias_value":"2608.00218v1","created_at":"2026-08-04T00:34:25.664986+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2608.00218","created_at":"2026-08-04T00:34:25.664986+00:00"},{"alias_kind":"pith_short_12","alias_value":"5BPXGGNB4ZRG","created_at":"2026-08-04T00:34:25.664986+00:00"},{"alias_kind":"pith_short_16","alias_value":"5BPXGGNB4ZRGLACO","created_at":"2026-08-04T00:34:25.664986+00:00"},{"alias_kind":"pith_short_8","alias_value":"5BPXGGNB","created_at":"2026-08-04T00:34:25.664986+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/5BPXGGNB4ZRGLACOBJVMZENMMZ","json":"https://pith.science/pith/5BPXGGNB4ZRGLACOBJVMZENMMZ.json","graph_json":"https://pith.science/api/pith-number/5BPXGGNB4ZRGLACOBJVMZENMMZ/graph.json","events_json":"https://pith.science/api/pith-number/5BPXGGNB4ZRGLACOBJVMZENMMZ/events.json","paper":"https://pith.science/paper/5BPXGGNB"},"agent_actions":{"view_html":"https://pith.science/pith/5BPXGGNB4ZRGLACOBJVMZENMMZ","download_json":"https://pith.science/pith/5BPXGGNB4ZRGLACOBJVMZENMMZ.json","view_paper":"https://pith.science/paper/5BPXGGNB","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2608.00218&json=true","fetch_graph":"https://pith.science/api/pith-number/5BPXGGNB4ZRGLACOBJVMZENMMZ/graph.json","fetch_events":"https://pith.science/api/pith-number/5BPXGGNB4ZRGLACOBJVMZENMMZ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/5BPXGGNB4ZRGLACOBJVMZENMMZ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/5BPXGGNB4ZRGLACOBJVMZENMMZ/action/storage_attestation","attest_author":"https://pith.science/pith/5BPXGGNB4ZRGLACOBJVMZENMMZ/action/author_attestation","sign_citation":"https://pith.science/pith/5BPXGGNB4ZRGLACOBJVMZENMMZ/action/citation_signature","submit_replication":"https://pith.science/pith/5BPXGGNB4ZRGLACOBJVMZENMMZ/action/replication_record"}},"created_at":"2026-08-04T00:34:25.664986+00:00","updated_at":"2026-08-04T00:34:25.664986+00:00"}