{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:OPJ54KWQNIIMTFTQTOSAVFUMAX","short_pith_number":"pith:OPJ54KWQ","schema_version":"1.0","canonical_sha256":"73d3de2ad06a10c996709ba40a968c05f2ac05b19667afd239c31295b087a721","source":{"kind":"arxiv","id":"2506.02048","version":2},"attestation_state":"computed","paper":{"title":"Improving LLM Agents with Reinforcement Learning on Cryptographic CTF Challenges","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CR","authors_text":"Andr\\'as Luk\\'acs, David Imolai, Lajos Muzsai","submitted_at":"2025-06-01T01:59:52Z","abstract_excerpt":"We present 'Random-Crypto', a procedurally generated cryptographic Capture The Flag (CTF) dataset designed to unlock the potential of Reinforcement Learning (RL) for LLM-based agents in security-sensitive domains. Cryptographic reasoning offers an ideal RL testbed: it combines precise validation, structured multi-step inference, and reliance on reliable computational tool use. Leveraging these properties, we fine-tune a Python tool-augmented Llama-3.1-8B via Group Relative Policy Optimization (GRPO) in a secure execution environment. The resulting agent achieves a significant improvement in Pa"},"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":"2506.02048","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.CR","submitted_at":"2025-06-01T01:59:52Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"3e4bbcf40d3d973d3ad5460481c4978efeb7a0db780715fcb75cb8149236d8c8","abstract_canon_sha256":"d129309e9cf94de362bc14a1ebc9a6e630e57a44416e3a512e4737887e4232da"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:54:47.816747Z","signature_b64":"qEqUX4160sxV+qqZqXKVxyy0cGaRS8KX7q3aI8mmuvYe9hUvvqsQ+TI3g2SnOxTNkoBIyxiuNZIcDU2l+3FZAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"73d3de2ad06a10c996709ba40a968c05f2ac05b19667afd239c31295b087a721","last_reissued_at":"2026-07-05T11:54:47.816241Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:54:47.816241Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Improving LLM Agents with Reinforcement Learning on Cryptographic CTF Challenges","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CR","authors_text":"Andr\\'as Luk\\'acs, David Imolai, Lajos Muzsai","submitted_at":"2025-06-01T01:59:52Z","abstract_excerpt":"We present 'Random-Crypto', a procedurally generated cryptographic Capture The Flag (CTF) dataset designed to unlock the potential of Reinforcement Learning (RL) for LLM-based agents in security-sensitive domains. Cryptographic reasoning offers an ideal RL testbed: it combines precise validation, structured multi-step inference, and reliance on reliable computational tool use. Leveraging these properties, we fine-tune a Python tool-augmented Llama-3.1-8B via Group Relative Policy Optimization (GRPO) in a secure execution environment. The resulting agent achieves a significant improvement in Pa"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.02048","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/2506.02048/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":"2506.02048","created_at":"2026-07-05T11:54:47.816299+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.02048v2","created_at":"2026-07-05T11:54:47.816299+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.02048","created_at":"2026-07-05T11:54:47.816299+00:00"},{"alias_kind":"pith_short_12","alias_value":"OPJ54KWQNIIM","created_at":"2026-07-05T11:54:47.816299+00:00"},{"alias_kind":"pith_short_16","alias_value":"OPJ54KWQNIIMTFTQ","created_at":"2026-07-05T11:54:47.816299+00:00"},{"alias_kind":"pith_short_8","alias_value":"OPJ54KWQ","created_at":"2026-07-05T11:54:47.816299+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.28146","citing_title":"Cybersecurity AI (CAI) Dataset","ref_index":45,"is_internal_anchor":false},{"citing_arxiv_id":"2602.05523","citing_title":"Capture the Flags: Family-Based Evaluation of Agentic LLMs via Semantics-Preserving Transformations","ref_index":33,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/OPJ54KWQNIIMTFTQTOSAVFUMAX","json":"https://pith.science/pith/OPJ54KWQNIIMTFTQTOSAVFUMAX.json","graph_json":"https://pith.science/api/pith-number/OPJ54KWQNIIMTFTQTOSAVFUMAX/graph.json","events_json":"https://pith.science/api/pith-number/OPJ54KWQNIIMTFTQTOSAVFUMAX/events.json","paper":"https://pith.science/paper/OPJ54KWQ"},"agent_actions":{"view_html":"https://pith.science/pith/OPJ54KWQNIIMTFTQTOSAVFUMAX","download_json":"https://pith.science/pith/OPJ54KWQNIIMTFTQTOSAVFUMAX.json","view_paper":"https://pith.science/paper/OPJ54KWQ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.02048&json=true","fetch_graph":"https://pith.science/api/pith-number/OPJ54KWQNIIMTFTQTOSAVFUMAX/graph.json","fetch_events":"https://pith.science/api/pith-number/OPJ54KWQNIIMTFTQTOSAVFUMAX/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/OPJ54KWQNIIMTFTQTOSAVFUMAX/action/timestamp_anchor","attest_storage":"https://pith.science/pith/OPJ54KWQNIIMTFTQTOSAVFUMAX/action/storage_attestation","attest_author":"https://pith.science/pith/OPJ54KWQNIIMTFTQTOSAVFUMAX/action/author_attestation","sign_citation":"https://pith.science/pith/OPJ54KWQNIIMTFTQTOSAVFUMAX/action/citation_signature","submit_replication":"https://pith.science/pith/OPJ54KWQNIIMTFTQTOSAVFUMAX/action/replication_record"}},"created_at":"2026-07-05T11:54:47.816299+00:00","updated_at":"2026-07-05T11:54:47.816299+00:00"}