{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:ESPUVMQI4KANVYGXVHXWIAXBOK","short_pith_number":"pith:ESPUVMQI","schema_version":"1.0","canonical_sha256":"249f4ab208e280dae0d7a9ef6402e172a53cce60cbe2eacaa4bc77b4d326d881","source":{"kind":"arxiv","id":"2311.09948","version":3},"attestation_state":"computed","paper":{"title":"Hijacking Large Language Models via Adversarial In-Context Learning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL","cs.CR"],"primary_cat":"cs.LG","authors_text":"Dongxiao Zhu, Prashant Khanduri, Saleh Zare Zade, Xiangyu Zhou, Yao Qiang","submitted_at":"2023-11-16T15:01:48Z","abstract_excerpt":"In-context learning (ICL) has emerged as a powerful paradigm leveraging LLMs for specific downstream tasks by utilizing labeled examples as demonstrations (demos) in the preconditioned prompts. Despite its promising performance, crafted adversarial attacks pose a notable threat to the robustness of LLMs. Existing attacks are either easy to detect, require a trigger in user input, or lack specificity towards ICL. To address these issues, this work introduces a novel transferable prompt injection attack against ICL, aiming to hijack LLMs to generate the target output or elicit harmful responses."},"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":"2311.09948","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2023-11-16T15:01:48Z","cross_cats_sorted":["cs.CL","cs.CR"],"title_canon_sha256":"bb4d604d3f0c3a6dbb91b95864aa31979ba634d86130ef78988002049dc0e756","abstract_canon_sha256":"4b9899d4e118a8ac8e8a6ba0399bdf9431855c1a7edb488042f98ab9d19c8eee"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:11:30.538980Z","signature_b64":"v7TraNdmPvyhfIrLLhUJ2WAf6KQ25J6u3VgZdsrL62SUkG1Jbjh1XPKcr5Bd4oZEEaHVjRVZYbnVNHyNo7A5Cw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"249f4ab208e280dae0d7a9ef6402e172a53cce60cbe2eacaa4bc77b4d326d881","last_reissued_at":"2026-07-05T11:11:30.538438Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:11:30.538438Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Hijacking Large Language Models via Adversarial In-Context Learning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL","cs.CR"],"primary_cat":"cs.LG","authors_text":"Dongxiao Zhu, Prashant Khanduri, Saleh Zare Zade, Xiangyu Zhou, Yao Qiang","submitted_at":"2023-11-16T15:01:48Z","abstract_excerpt":"In-context learning (ICL) has emerged as a powerful paradigm leveraging LLMs for specific downstream tasks by utilizing labeled examples as demonstrations (demos) in the preconditioned prompts. Despite its promising performance, crafted adversarial attacks pose a notable threat to the robustness of LLMs. Existing attacks are either easy to detect, require a trigger in user input, or lack specificity towards ICL. To address these issues, this work introduces a novel transferable prompt injection attack against ICL, aiming to hijack LLMs to generate the target output or elicit harmful responses."},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2311.09948","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/2311.09948/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":"2311.09948","created_at":"2026-07-05T11:11:30.538506+00:00"},{"alias_kind":"arxiv_version","alias_value":"2311.09948v3","created_at":"2026-07-05T11:11:30.538506+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2311.09948","created_at":"2026-07-05T11:11:30.538506+00:00"},{"alias_kind":"pith_short_12","alias_value":"ESPUVMQI4KAN","created_at":"2026-07-05T11:11:30.538506+00:00"},{"alias_kind":"pith_short_16","alias_value":"ESPUVMQI4KANVYGX","created_at":"2026-07-05T11:11:30.538506+00:00"},{"alias_kind":"pith_short_8","alias_value":"ESPUVMQI","created_at":"2026-07-05T11:11:30.538506+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":6,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.26350","citing_title":"When Correct Demonstrations Hurt: Rethinking the Role of Exemplars in In-Context Learning","ref_index":18,"is_internal_anchor":false},{"citing_arxiv_id":"2601.17887","citing_title":"When Personalization Legitimizes Risks: Uncovering Safety Vulnerabilities in Personalized Dialogue Agents","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12813","citing_title":"REALISTA: Realistic Latent Adversarial Attacks that Elicit LLM Hallucinations","ref_index":93,"is_internal_anchor":false},{"citing_arxiv_id":"2605.05116","citing_title":"On the Hardness of Junking LLMs","ref_index":64,"is_internal_anchor":false},{"citing_arxiv_id":"2502.18864","citing_title":"Towards an AI co-scientist","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2604.23593","citing_title":"When AI reviews science: Can we trust the referee?","ref_index":108,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/ESPUVMQI4KANVYGXVHXWIAXBOK","json":"https://pith.science/pith/ESPUVMQI4KANVYGXVHXWIAXBOK.json","graph_json":"https://pith.science/api/pith-number/ESPUVMQI4KANVYGXVHXWIAXBOK/graph.json","events_json":"https://pith.science/api/pith-number/ESPUVMQI4KANVYGXVHXWIAXBOK/events.json","paper":"https://pith.science/paper/ESPUVMQI"},"agent_actions":{"view_html":"https://pith.science/pith/ESPUVMQI4KANVYGXVHXWIAXBOK","download_json":"https://pith.science/pith/ESPUVMQI4KANVYGXVHXWIAXBOK.json","view_paper":"https://pith.science/paper/ESPUVMQI","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2311.09948&json=true","fetch_graph":"https://pith.science/api/pith-number/ESPUVMQI4KANVYGXVHXWIAXBOK/graph.json","fetch_events":"https://pith.science/api/pith-number/ESPUVMQI4KANVYGXVHXWIAXBOK/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ESPUVMQI4KANVYGXVHXWIAXBOK/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ESPUVMQI4KANVYGXVHXWIAXBOK/action/storage_attestation","attest_author":"https://pith.science/pith/ESPUVMQI4KANVYGXVHXWIAXBOK/action/author_attestation","sign_citation":"https://pith.science/pith/ESPUVMQI4KANVYGXVHXWIAXBOK/action/citation_signature","submit_replication":"https://pith.science/pith/ESPUVMQI4KANVYGXVHXWIAXBOK/action/replication_record"}},"created_at":"2026-07-05T11:11:30.538506+00:00","updated_at":"2026-07-05T11:11:30.538506+00:00"}