{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:3ATN6BEHMK3KS46RDWZWV4WA7B","short_pith_number":"pith:3ATN6BEH","schema_version":"1.0","canonical_sha256":"d826df048762b6a973d11db36af2c0f852602b13484f0f2ac16523cc1238599c","source":{"kind":"arxiv","id":"2311.01011","version":1},"attestation_state":"computed","paper":{"title":"Tensor Trust: Interpretable Prompt Injection Attacks from an Online Game","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CR"],"primary_cat":"cs.LG","authors_text":"Alan Ritter, Ethan Adrian Mendes, Isaac Ong, Justin Svegliato, Karim Elmaaroufi, Luke Bailey, Olivia Watkins, Pieter Abbeel, Sam Toyer, Stuart Russell, Tiffany Wang, Trevor Darrell","submitted_at":"2023-11-02T06:13:36Z","abstract_excerpt":"While Large Language Models (LLMs) are increasingly being used in real-world applications, they remain vulnerable to prompt injection attacks: malicious third party prompts that subvert the intent of the system designer. To help researchers study this problem, we present a dataset of over 126,000 prompt injection attacks and 46,000 prompt-based \"defenses\" against prompt injection, all created by players of an online game called Tensor Trust. To the best of our knowledge, this is currently the largest dataset of human-generated adversarial examples for instruction-following LLMs. The attacks in"},"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":"2311.01011","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2023-11-02T06:13:36Z","cross_cats_sorted":["cs.CR"],"title_canon_sha256":"597da5cfa5f0c63eb49ec395866676c5cba81c2273b1c09fe733c551db2bc423","abstract_canon_sha256":"ae7d3c4bc0e43c9c4224a8a3bcb9e526f59254d112e2f3d2714f9ac0d6414917"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:08:19.246396Z","signature_b64":"9PdyaoJ+Fd4ZY8gDnyN9jx+kLDlEDEb6fJf7Lju8cwWG6AEMo0HmAgAxKhKfoZqXKSBj6Zs6CsxekY4PYYMtDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"d826df048762b6a973d11db36af2c0f852602b13484f0f2ac16523cc1238599c","last_reissued_at":"2026-07-05T07:08:19.245902Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:08:19.245902Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Tensor Trust: Interpretable Prompt Injection Attacks from an Online Game","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CR"],"primary_cat":"cs.LG","authors_text":"Alan Ritter, Ethan Adrian Mendes, Isaac Ong, Justin Svegliato, Karim Elmaaroufi, Luke Bailey, Olivia Watkins, Pieter Abbeel, Sam Toyer, Stuart Russell, Tiffany Wang, Trevor Darrell","submitted_at":"2023-11-02T06:13:36Z","abstract_excerpt":"While Large Language Models (LLMs) are increasingly being used in real-world applications, they remain vulnerable to prompt injection attacks: malicious third party prompts that subvert the intent of the system designer. To help researchers study this problem, we present a dataset of over 126,000 prompt injection attacks and 46,000 prompt-based \"defenses\" against prompt injection, all created by players of an online game called Tensor Trust. To the best of our knowledge, this is currently the largest dataset of human-generated adversarial examples for instruction-following LLMs. The attacks in"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2311.01011","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/2311.01011/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":"2311.01011","created_at":"2026-07-05T07:08:19.245962+00:00"},{"alias_kind":"arxiv_version","alias_value":"2311.01011v1","created_at":"2026-07-05T07:08:19.245962+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2311.01011","created_at":"2026-07-05T07:08:19.245962+00:00"},{"alias_kind":"pith_short_12","alias_value":"3ATN6BEHMK3K","created_at":"2026-07-05T07:08:19.245962+00:00"},{"alias_kind":"pith_short_16","alias_value":"3ATN6BEHMK3KS46R","created_at":"2026-07-05T07:08:19.245962+00:00"},{"alias_kind":"pith_short_8","alias_value":"3ATN6BEH","created_at":"2026-07-05T07:08:19.245962+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":15,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.04141","citing_title":"Caught in the Act(ivation): Toward Pre-Output and Multi-Turn Detection of Credential Exfiltration by LLM Agents","ref_index":16,"is_internal_anchor":false},{"citing_arxiv_id":"2605.29251","citing_title":"Provably Secure Agent Guardrail","ref_index":50,"is_internal_anchor":false},{"citing_arxiv_id":"2605.30521","citing_title":"Evaluating using Mock Tool Calls to Quarantine Untrusted Prompt Inputs","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2403.10559","citing_title":"Generative Models and Connected and Automated Vehicles: A Survey in Exploring the Intersection of Transportation and AI","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2604.23887","citing_title":"Evaluation of Prompt Injection Defenses in Large Language Models","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2605.13471","citing_title":"Sleeper Channels and Provenance Gates: Persistent Prompt Injection in Always-on Autonomous AI Agents","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2406.13352","citing_title":"AgentDojo: A Dynamic Environment to Evaluate Prompt Injection Attacks and Defenses for LLM Agents","ref_index":57,"is_internal_anchor":false},{"citing_arxiv_id":"2605.11868","citing_title":"IPI-proxy: An Intercepting Proxy for Red-Teaming Web-Browsing AI Agents Against Indirect Prompt Injection","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2410.02644","citing_title":"Agent Security Bench (ASB): Formalizing and Benchmarking Attacks and Defenses in LLM-based Agents","ref_index":134,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08646","citing_title":"PAAC: Privacy-Aware Agentic Device-Cloud Collaboration","ref_index":42,"is_internal_anchor":false},{"citing_arxiv_id":"2604.24826","citing_title":"A Comparative Evaluation of AI Agent Security Guardrails","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2604.23887","citing_title":"Evaluation of Prompt Injection Defenses in Large Language Models","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2604.10326","citing_title":"Jailbreaking the Matrix: Nullspace Steering for Controlled Model Subversion","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2604.06247","citing_title":"SALLIE: Safeguarding Against Latent Language & Image Exploits","ref_index":26,"is_internal_anchor":false},{"citing_arxiv_id":"2604.17396","citing_title":"Representation-Guided Parameter-Efficient LLM Unlearning","ref_index":45,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/3ATN6BEHMK3KS46RDWZWV4WA7B","json":"https://pith.science/pith/3ATN6BEHMK3KS46RDWZWV4WA7B.json","graph_json":"https://pith.science/api/pith-number/3ATN6BEHMK3KS46RDWZWV4WA7B/graph.json","events_json":"https://pith.science/api/pith-number/3ATN6BEHMK3KS46RDWZWV4WA7B/events.json","paper":"https://pith.science/paper/3ATN6BEH"},"agent_actions":{"view_html":"https://pith.science/pith/3ATN6BEHMK3KS46RDWZWV4WA7B","download_json":"https://pith.science/pith/3ATN6BEHMK3KS46RDWZWV4WA7B.json","view_paper":"https://pith.science/paper/3ATN6BEH","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2311.01011&json=true","fetch_graph":"https://pith.science/api/pith-number/3ATN6BEHMK3KS46RDWZWV4WA7B/graph.json","fetch_events":"https://pith.science/api/pith-number/3ATN6BEHMK3KS46RDWZWV4WA7B/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/3ATN6BEHMK3KS46RDWZWV4WA7B/action/timestamp_anchor","attest_storage":"https://pith.science/pith/3ATN6BEHMK3KS46RDWZWV4WA7B/action/storage_attestation","attest_author":"https://pith.science/pith/3ATN6BEHMK3KS46RDWZWV4WA7B/action/author_attestation","sign_citation":"https://pith.science/pith/3ATN6BEHMK3KS46RDWZWV4WA7B/action/citation_signature","submit_replication":"https://pith.science/pith/3ATN6BEHMK3KS46RDWZWV4WA7B/action/replication_record"}},"created_at":"2026-07-05T07:08:19.245962+00:00","updated_at":"2026-07-05T07:08:19.245962+00:00"}