{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:XQXLJV74NXCYXWH36VKAM4WBRE","short_pith_number":"pith:XQXLJV74","schema_version":"1.0","canonical_sha256":"bc2eb4d7fc6dc58bd8fbf5540672c18904d9e8703d48f85319c5d7fc759b2df2","source":{"kind":"arxiv","id":"2606.07867","version":1},"attestation_state":"computed","paper":{"title":"The Cold-Start Safety Gap in LLM Agents","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Chung-En Sun, Linbo Liu, Tsui-Wei Weng","submitted_at":"2026-06-05T21:58:19Z","abstract_excerpt":"Are tool-calling LLM agents equally safe throughout a conversation? We discover they are not: agents are most vulnerable at the very start of a session and become substantially safer after a few regular agentic tasks -- a phenomenon we term the cold-start safety gap. To study this systematically, we introduce Safety Over Depth for Agents (SODA), a benchmark that controls how many regular agentic tasks the agent completes before encountering a safety threat, supporting up to 20 preceding tasks. Evaluating 7 models from 4 families, safety improves by 9--52% as the number of preceding regular age"},"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":"2606.07867","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2026-06-05T21:58:19Z","cross_cats_sorted":[],"title_canon_sha256":"97d490be9f755cbead2d167f98ead0da4c6ab8d6837754f145f1e53a4c967a53","abstract_canon_sha256":"1077ed1f026ef71bd9aca4d5702111e37af6bd48bf23324324f3a078f99cef3c"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-09T01:04:54.149634Z","signature_b64":"L1lfwP3wVOTjPfi4mlbvEjmCbD1XpXQA+zVPJyJeYMXyg3sSDzBXWwIfDca5I9W7G+gHewIgl/2wxSw7MihQBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"bc2eb4d7fc6dc58bd8fbf5540672c18904d9e8703d48f85319c5d7fc759b2df2","last_reissued_at":"2026-06-09T01:04:54.149222Z","signature_status":"signed_v1","first_computed_at":"2026-06-09T01:04:54.149222Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"The Cold-Start Safety Gap in LLM Agents","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Chung-En Sun, Linbo Liu, Tsui-Wei Weng","submitted_at":"2026-06-05T21:58:19Z","abstract_excerpt":"Are tool-calling LLM agents equally safe throughout a conversation? We discover they are not: agents are most vulnerable at the very start of a session and become substantially safer after a few regular agentic tasks -- a phenomenon we term the cold-start safety gap. To study this systematically, we introduce Safety Over Depth for Agents (SODA), a benchmark that controls how many regular agentic tasks the agent completes before encountering a safety threat, supporting up to 20 preceding tasks. Evaluating 7 models from 4 families, safety improves by 9--52% as the number of preceding regular age"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.07867","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/2606.07867/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":"2606.07867","created_at":"2026-06-09T01:04:54.149283+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.07867v1","created_at":"2026-06-09T01:04:54.149283+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.07867","created_at":"2026-06-09T01:04:54.149283+00:00"},{"alias_kind":"pith_short_12","alias_value":"XQXLJV74NXCY","created_at":"2026-06-09T01:04:54.149283+00:00"},{"alias_kind":"pith_short_16","alias_value":"XQXLJV74NXCYXWH3","created_at":"2026-06-09T01:04:54.149283+00:00"},{"alias_kind":"pith_short_8","alias_value":"XQXLJV74","created_at":"2026-06-09T01:04:54.149283+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/XQXLJV74NXCYXWH36VKAM4WBRE","json":"https://pith.science/pith/XQXLJV74NXCYXWH36VKAM4WBRE.json","graph_json":"https://pith.science/api/pith-number/XQXLJV74NXCYXWH36VKAM4WBRE/graph.json","events_json":"https://pith.science/api/pith-number/XQXLJV74NXCYXWH36VKAM4WBRE/events.json","paper":"https://pith.science/paper/XQXLJV74"},"agent_actions":{"view_html":"https://pith.science/pith/XQXLJV74NXCYXWH36VKAM4WBRE","download_json":"https://pith.science/pith/XQXLJV74NXCYXWH36VKAM4WBRE.json","view_paper":"https://pith.science/paper/XQXLJV74","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.07867&json=true","fetch_graph":"https://pith.science/api/pith-number/XQXLJV74NXCYXWH36VKAM4WBRE/graph.json","fetch_events":"https://pith.science/api/pith-number/XQXLJV74NXCYXWH36VKAM4WBRE/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/XQXLJV74NXCYXWH36VKAM4WBRE/action/timestamp_anchor","attest_storage":"https://pith.science/pith/XQXLJV74NXCYXWH36VKAM4WBRE/action/storage_attestation","attest_author":"https://pith.science/pith/XQXLJV74NXCYXWH36VKAM4WBRE/action/author_attestation","sign_citation":"https://pith.science/pith/XQXLJV74NXCYXWH36VKAM4WBRE/action/citation_signature","submit_replication":"https://pith.science/pith/XQXLJV74NXCYXWH36VKAM4WBRE/action/replication_record"}},"created_at":"2026-06-09T01:04:54.149283+00:00","updated_at":"2026-06-09T01:04:54.149283+00:00"}