{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:DPIJIZRQ6TLRXHVMLDSRPSL2NA","short_pith_number":"pith:DPIJIZRQ","schema_version":"1.0","canonical_sha256":"1bd0946630f4d71b9eac58e517c97a682ffda0a8d657f401afb7f7ce6c075878","source":{"kind":"arxiv","id":"2509.07617","version":1},"attestation_state":"computed","paper":{"title":"Transferable Direct Prompt Injection via Activation-Guided MCMC Sampling","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Hao Zhang, Jing Wang, Minghui Li, Pei Xiaobing, Shengshan Hu, Wei Wan, Yechao Zhang","submitted_at":"2025-09-09T11:42:06Z","abstract_excerpt":"Direct Prompt Injection (DPI) attacks pose a critical security threat to Large Language Models (LLMs) due to their low barrier of execution and high potential damage. To address the impracticality of existing white-box/gray-box methods and the poor transferability of black-box methods, we propose an activations-guided prompt injection attack framework. We first construct an Energy-based Model (EBM) using activations from a surrogate model to evaluate the quality of adversarial prompts. Guided by the trained EBM, we employ the token-level Markov Chain Monte Carlo (MCMC) sampling to adaptively o"},"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":"2509.07617","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2025-09-09T11:42:06Z","cross_cats_sorted":[],"title_canon_sha256":"ebecec5787b2a4836a2e4100c763391351f7c5d2f8cd352ae78002c033ed1b3c","abstract_canon_sha256":"790c288f003f671872ecccf689e7ad0941a83fc4b220e8d1a9980926b955ae29"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T12:07:24.756298Z","signature_b64":"eAOkuXeM1ag8JMfQfOAZ6sWrROcJ3PziKhu6hZrR5BePUXw1ownz4jSXXhOLWIyBeJEYDrhTSQYtVI7bOJupCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"1bd0946630f4d71b9eac58e517c97a682ffda0a8d657f401afb7f7ce6c075878","last_reissued_at":"2026-07-05T12:07:24.755876Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T12:07:24.755876Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Transferable Direct Prompt Injection via Activation-Guided MCMC Sampling","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Hao Zhang, Jing Wang, Minghui Li, Pei Xiaobing, Shengshan Hu, Wei Wan, Yechao Zhang","submitted_at":"2025-09-09T11:42:06Z","abstract_excerpt":"Direct Prompt Injection (DPI) attacks pose a critical security threat to Large Language Models (LLMs) due to their low barrier of execution and high potential damage. To address the impracticality of existing white-box/gray-box methods and the poor transferability of black-box methods, we propose an activations-guided prompt injection attack framework. We first construct an Energy-based Model (EBM) using activations from a surrogate model to evaluate the quality of adversarial prompts. Guided by the trained EBM, we employ the token-level Markov Chain Monte Carlo (MCMC) sampling to adaptively o"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2509.07617","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/2509.07617/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":"2509.07617","created_at":"2026-07-05T12:07:24.755929+00:00"},{"alias_kind":"arxiv_version","alias_value":"2509.07617v1","created_at":"2026-07-05T12:07:24.755929+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2509.07617","created_at":"2026-07-05T12:07:24.755929+00:00"},{"alias_kind":"pith_short_12","alias_value":"DPIJIZRQ6TLR","created_at":"2026-07-05T12:07:24.755929+00:00"},{"alias_kind":"pith_short_16","alias_value":"DPIJIZRQ6TLRXHVM","created_at":"2026-07-05T12:07:24.755929+00:00"},{"alias_kind":"pith_short_8","alias_value":"DPIJIZRQ","created_at":"2026-07-05T12:07:24.755929+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/DPIJIZRQ6TLRXHVMLDSRPSL2NA","json":"https://pith.science/pith/DPIJIZRQ6TLRXHVMLDSRPSL2NA.json","graph_json":"https://pith.science/api/pith-number/DPIJIZRQ6TLRXHVMLDSRPSL2NA/graph.json","events_json":"https://pith.science/api/pith-number/DPIJIZRQ6TLRXHVMLDSRPSL2NA/events.json","paper":"https://pith.science/paper/DPIJIZRQ"},"agent_actions":{"view_html":"https://pith.science/pith/DPIJIZRQ6TLRXHVMLDSRPSL2NA","download_json":"https://pith.science/pith/DPIJIZRQ6TLRXHVMLDSRPSL2NA.json","view_paper":"https://pith.science/paper/DPIJIZRQ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2509.07617&json=true","fetch_graph":"https://pith.science/api/pith-number/DPIJIZRQ6TLRXHVMLDSRPSL2NA/graph.json","fetch_events":"https://pith.science/api/pith-number/DPIJIZRQ6TLRXHVMLDSRPSL2NA/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/DPIJIZRQ6TLRXHVMLDSRPSL2NA/action/timestamp_anchor","attest_storage":"https://pith.science/pith/DPIJIZRQ6TLRXHVMLDSRPSL2NA/action/storage_attestation","attest_author":"https://pith.science/pith/DPIJIZRQ6TLRXHVMLDSRPSL2NA/action/author_attestation","sign_citation":"https://pith.science/pith/DPIJIZRQ6TLRXHVMLDSRPSL2NA/action/citation_signature","submit_replication":"https://pith.science/pith/DPIJIZRQ6TLRXHVMLDSRPSL2NA/action/replication_record"}},"created_at":"2026-07-05T12:07:24.755929+00:00","updated_at":"2026-07-05T12:07:24.755929+00:00"}