{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:4IOU7DUOOR2FSEEDNKMTI5GLEP","short_pith_number":"pith:4IOU7DUO","schema_version":"1.0","canonical_sha256":"e21d4f8e8e74745910836a993474cb23c4a347b5947d9c176223f8d495587ba8","source":{"kind":"arxiv","id":"2411.05189","version":2},"attestation_state":"computed","paper":{"title":"Understanding In-Context Learning of Linear Models in Transformers Through an Adversarial Lens","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CR"],"primary_cat":"cs.LG","authors_text":"David Krueger, Johannes von Oswald, Louis Kirsch, Spencer Frei, Usman Anwar","submitted_at":"2024-11-07T21:25:58Z","abstract_excerpt":"In this work, we make two contributions towards understanding of in-context learning of linear models by transformers. First, we investigate the adversarial robustness of in-context learning in transformers to hijacking attacks -- a type of adversarial attacks in which the adversary's goal is to manipulate the prompt to force the transformer to generate a specific output. We show that both linear transformers and transformers with GPT-2 architectures are vulnerable to such hijacking attacks. However, adversarial robustness to such attacks can be significantly improved through adversarial train"},"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":"2411.05189","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-11-07T21:25:58Z","cross_cats_sorted":["cs.CR"],"title_canon_sha256":"451301b5fae42729d876b81f0e212443076b866b516a47782b3d7793ce861e18","abstract_canon_sha256":"cc8e7ef0150925a2b27aca6982afd6c82e4a033319b5f920ffaa7bebd8319dff"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:49:05.187233Z","signature_b64":"qKlGSu746BahTRc3mz0yecLpEKEjzUC4pcVLde0M/F2/1kVDD3IesWK6EkCi2fpWZyYw+eoPlyhJJODd5+6oCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"e21d4f8e8e74745910836a993474cb23c4a347b5947d9c176223f8d495587ba8","last_reissued_at":"2026-07-05T11:49:05.186800Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:49:05.186800Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Understanding In-Context Learning of Linear Models in Transformers Through an Adversarial Lens","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CR"],"primary_cat":"cs.LG","authors_text":"David Krueger, Johannes von Oswald, Louis Kirsch, Spencer Frei, Usman Anwar","submitted_at":"2024-11-07T21:25:58Z","abstract_excerpt":"In this work, we make two contributions towards understanding of in-context learning of linear models by transformers. First, we investigate the adversarial robustness of in-context learning in transformers to hijacking attacks -- a type of adversarial attacks in which the adversary's goal is to manipulate the prompt to force the transformer to generate a specific output. We show that both linear transformers and transformers with GPT-2 architectures are vulnerable to such hijacking attacks. However, adversarial robustness to such attacks can be significantly improved through adversarial train"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2411.05189","kind":"arxiv","version":2},"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/2411.05189/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":"2411.05189","created_at":"2026-07-05T11:49:05.186859+00:00"},{"alias_kind":"arxiv_version","alias_value":"2411.05189v2","created_at":"2026-07-05T11:49:05.186859+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2411.05189","created_at":"2026-07-05T11:49:05.186859+00:00"},{"alias_kind":"pith_short_12","alias_value":"4IOU7DUOOR2F","created_at":"2026-07-05T11:49:05.186859+00:00"},{"alias_kind":"pith_short_16","alias_value":"4IOU7DUOOR2FSEED","created_at":"2026-07-05T11:49:05.186859+00:00"},{"alias_kind":"pith_short_8","alias_value":"4IOU7DUO","created_at":"2026-07-05T11:49:05.186859+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2506.02978","citing_title":"On the Robustness of Tabular Foundation Models: Test-Time Attacks and In-Context Defenses","ref_index":1,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06609","citing_title":"Transformers Efficiently Perform In-Context Logistic Regression via Normalized Gradient Descent","ref_index":3,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/4IOU7DUOOR2FSEEDNKMTI5GLEP","json":"https://pith.science/pith/4IOU7DUOOR2FSEEDNKMTI5GLEP.json","graph_json":"https://pith.science/api/pith-number/4IOU7DUOOR2FSEEDNKMTI5GLEP/graph.json","events_json":"https://pith.science/api/pith-number/4IOU7DUOOR2FSEEDNKMTI5GLEP/events.json","paper":"https://pith.science/paper/4IOU7DUO"},"agent_actions":{"view_html":"https://pith.science/pith/4IOU7DUOOR2FSEEDNKMTI5GLEP","download_json":"https://pith.science/pith/4IOU7DUOOR2FSEEDNKMTI5GLEP.json","view_paper":"https://pith.science/paper/4IOU7DUO","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2411.05189&json=true","fetch_graph":"https://pith.science/api/pith-number/4IOU7DUOOR2FSEEDNKMTI5GLEP/graph.json","fetch_events":"https://pith.science/api/pith-number/4IOU7DUOOR2FSEEDNKMTI5GLEP/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/4IOU7DUOOR2FSEEDNKMTI5GLEP/action/timestamp_anchor","attest_storage":"https://pith.science/pith/4IOU7DUOOR2FSEEDNKMTI5GLEP/action/storage_attestation","attest_author":"https://pith.science/pith/4IOU7DUOOR2FSEEDNKMTI5GLEP/action/author_attestation","sign_citation":"https://pith.science/pith/4IOU7DUOOR2FSEEDNKMTI5GLEP/action/citation_signature","submit_replication":"https://pith.science/pith/4IOU7DUOOR2FSEEDNKMTI5GLEP/action/replication_record"}},"created_at":"2026-07-05T11:49:05.186859+00:00","updated_at":"2026-07-05T11:49:05.186859+00:00"}