{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:MKQPM4ZMATTQLKSWLTQLUGNNVJ","short_pith_number":"pith:MKQPM4ZM","schema_version":"1.0","canonical_sha256":"62a0f6732c04e705aa565ce0ba19adaa775f86fbb3ef5b07c1831f2a0973726e","source":{"kind":"arxiv","id":"2311.07590","version":4},"attestation_state":"computed","paper":{"title":"Large Language Models can Strategically Deceive their Users when Put Under Pressure","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"J\\'er\\'emy Scheurer, Marius Hobbhahn, Mikita Balesni","submitted_at":"2023-11-09T17:12:44Z","abstract_excerpt":"We demonstrate a situation in which Large Language Models, trained to be helpful, harmless, and honest, can display misaligned behavior and strategically deceive their users about this behavior without being instructed to do so. Concretely, we deploy GPT-4 as an agent in a realistic, simulated environment, where it assumes the role of an autonomous stock trading agent. Within this environment, the model obtains an insider tip about a lucrative stock trade and acts upon it despite knowing that insider trading is disapproved of by company management. When reporting to its manager, the model cons"},"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":"2311.07590","kind":"arxiv","version":4},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2023-11-09T17:12:44Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"0b2f87701aa14b050fec958c823b3c8f3382061395028a0a4f7dc722cbf1ed48","abstract_canon_sha256":"fcb37fa4b30038eb2271a434491fdb7f1581c8ba9a2e134418be8dc13165d318"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:43:51.445771Z","signature_b64":"EcfjfxFJd6W82m1LhqWVoZUDNaxoBkrbp7ypBWczt09VxfvirXb7nnhZAuEp47/RuD8WVvaQGxd9EyWkYR2nAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"62a0f6732c04e705aa565ce0ba19adaa775f86fbb3ef5b07c1831f2a0973726e","last_reissued_at":"2026-07-05T08:43:51.445287Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:43:51.445287Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Large Language Models can Strategically Deceive their Users when Put Under Pressure","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"J\\'er\\'emy Scheurer, Marius Hobbhahn, Mikita Balesni","submitted_at":"2023-11-09T17:12:44Z","abstract_excerpt":"We demonstrate a situation in which Large Language Models, trained to be helpful, harmless, and honest, can display misaligned behavior and strategically deceive their users about this behavior without being instructed to do so. Concretely, we deploy GPT-4 as an agent in a realistic, simulated environment, where it assumes the role of an autonomous stock trading agent. Within this environment, the model obtains an insider tip about a lucrative stock trade and acts upon it despite knowing that insider trading is disapproved of by company management. When reporting to its manager, the model cons"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2311.07590","kind":"arxiv","version":4},"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/2311.07590/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":"2311.07590","created_at":"2026-07-05T08:43:51.445359+00:00"},{"alias_kind":"arxiv_version","alias_value":"2311.07590v4","created_at":"2026-07-05T08:43:51.445359+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2311.07590","created_at":"2026-07-05T08:43:51.445359+00:00"},{"alias_kind":"pith_short_12","alias_value":"MKQPM4ZMATTQ","created_at":"2026-07-05T08:43:51.445359+00:00"},{"alias_kind":"pith_short_16","alias_value":"MKQPM4ZMATTQLKSW","created_at":"2026-07-05T08:43:51.445359+00:00"},{"alias_kind":"pith_short_8","alias_value":"MKQPM4ZM","created_at":"2026-07-05T08:43:51.445359+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":18,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.25899","citing_title":"Manipulation Is Task-Dependent: A Multi-Axis, Multi-Environment Evaluation of Frontier LLMs","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2606.31916","citing_title":"Theory of Mind and Persuasion Beyond Conversation: Assessing the Capacity of LLMs to Induce Belief States via Planning and Action","ref_index":64,"is_internal_anchor":false},{"citing_arxiv_id":"2605.28114","citing_title":"Human-like in-group bias in instruction-tuned language model agents","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2605.09391","citing_title":"Do Linear Probes Generalize Better in Persona Coordinates?","ref_index":64,"is_internal_anchor":false},{"citing_arxiv_id":"2509.06701","citing_title":"Probabilistic Modeling of Latent Agentic Substructures in Deep Neural Networks","ref_index":24,"is_internal_anchor":false},{"citing_arxiv_id":"2511.17408","citing_title":"The Impact of Off-Policy Training Data on Probe Generalisation","ref_index":33,"is_internal_anchor":false},{"citing_arxiv_id":"2406.10162","citing_title":"Sycophancy to Subterfuge: Investigating Reward-Tampering in Large Language Models","ref_index":95,"is_internal_anchor":false},{"citing_arxiv_id":"2605.13329","citing_title":"Tracing Persona Vectors Through LLM Pretraining","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2604.01151","citing_title":"Detecting Multi-Agent Collusion Through Multi-Agent Interpretability","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2604.03121","citing_title":"An Independent Safety Evaluation of Kimi K2.5","ref_index":48,"is_internal_anchor":false},{"citing_arxiv_id":"2604.04157","citing_title":"Readable Minds: Emergent Theory-of-Mind-Like Behavior in LLM Poker Agents","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2605.11448","citing_title":"Deep Minds and Shallow Probes","ref_index":39,"is_internal_anchor":false},{"citing_arxiv_id":"2605.09391","citing_title":"Do Linear Probes Generalize Better in Persona Coordinates?","ref_index":65,"is_internal_anchor":false},{"citing_arxiv_id":"2604.24966","citing_title":"Risk Reporting for Developers' Internal AI Model Use","ref_index":40,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06490","citing_title":"Instrumental Choices: Measuring the Propensity of LLM Agents to Pursue Instrumental Behaviors","ref_index":26,"is_internal_anchor":false},{"citing_arxiv_id":"2604.09104","citing_title":"Scheming in the wild: detecting real-world AI scheming incidents with open-source intelligence","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2604.04561","citing_title":"Mapping the Exploitation Surface: A 10,000-Trial Taxonomy of What Makes LLM Agents Exploit Vulnerabilities","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06327","citing_title":"Measuring Evaluation-Context Divergence in Open-Weight LLMs: A Paired-Prompt Protocol with Pilot Evidence of Alignment-Pipeline-Specific Heterogeneity","ref_index":42,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/MKQPM4ZMATTQLKSWLTQLUGNNVJ","json":"https://pith.science/pith/MKQPM4ZMATTQLKSWLTQLUGNNVJ.json","graph_json":"https://pith.science/api/pith-number/MKQPM4ZMATTQLKSWLTQLUGNNVJ/graph.json","events_json":"https://pith.science/api/pith-number/MKQPM4ZMATTQLKSWLTQLUGNNVJ/events.json","paper":"https://pith.science/paper/MKQPM4ZM"},"agent_actions":{"view_html":"https://pith.science/pith/MKQPM4ZMATTQLKSWLTQLUGNNVJ","download_json":"https://pith.science/pith/MKQPM4ZMATTQLKSWLTQLUGNNVJ.json","view_paper":"https://pith.science/paper/MKQPM4ZM","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2311.07590&json=true","fetch_graph":"https://pith.science/api/pith-number/MKQPM4ZMATTQLKSWLTQLUGNNVJ/graph.json","fetch_events":"https://pith.science/api/pith-number/MKQPM4ZMATTQLKSWLTQLUGNNVJ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/MKQPM4ZMATTQLKSWLTQLUGNNVJ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/MKQPM4ZMATTQLKSWLTQLUGNNVJ/action/storage_attestation","attest_author":"https://pith.science/pith/MKQPM4ZMATTQLKSWLTQLUGNNVJ/action/author_attestation","sign_citation":"https://pith.science/pith/MKQPM4ZMATTQLKSWLTQLUGNNVJ/action/citation_signature","submit_replication":"https://pith.science/pith/MKQPM4ZMATTQLKSWLTQLUGNNVJ/action/replication_record"}},"created_at":"2026-07-05T08:43:51.445359+00:00","updated_at":"2026-07-05T08:43:51.445359+00:00"}