{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:JDBNSSL7M7EKJGGT55HPZCN7HP","short_pith_number":"pith:JDBNSSL7","schema_version":"1.0","canonical_sha256":"48c2d9497f67c8a498d3ef4efc89bf3be0f90480ed2852718c5ea7883f519d39","source":{"kind":"arxiv","id":"2606.08960","version":1},"attestation_state":"computed","paper":{"title":"Hardening Agent Benchmarks with Adversarial Hacker-Fixer Loops","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG","cs.MA"],"primary_cat":"cs.CR","authors_text":"Aditi Raghunathan, Ivan Bercovich, Ivgeni Segal, Kexun Zhang, Shashwat Saxena, Ziqian Zhong","submitted_at":"2026-06-08T03:00:56Z","abstract_excerpt":"Agent benchmarks score submissions with outcome verifiers that are typically hand-written and brittle, leaving them open to reward hacking. We audit 1,968 tasks across five terminal-agent benchmarks and find 323 (16%) hackable by frontier models given only the task description. This corrupts both leaderboard rankings and RL training signal, yet the standard response is manual and reactive.\n  We introduce the hacker-fixer loop, a method for building exploit-resistant verifiers without per-task manual patching. The loop alternates three LLM agents: a hacker tries to pass the verifier without sol"},"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":"2606.08960","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CR","submitted_at":"2026-06-08T03:00:56Z","cross_cats_sorted":["cs.AI","cs.LG","cs.MA"],"title_canon_sha256":"84c30a74b34596d006804b8581b935d59b40e89bdc2b99711ea453c05dcaef80","abstract_canon_sha256":"3abf44ebf17ca669d8740e7a0533d4962b8f775b3e211a32d72a9b990347a6d9"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-09T02:07:50.395629Z","signature_b64":"FZRmUk/ebkcT2zMA8CQ+9HBEhaqpqaHv9JdNKrMUeOYKYDhXunD84Axg4tzZKr9HGdCwX/x9vo6SyMjEr4oSBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"48c2d9497f67c8a498d3ef4efc89bf3be0f90480ed2852718c5ea7883f519d39","last_reissued_at":"2026-06-09T02:07:50.394741Z","signature_status":"signed_v1","first_computed_at":"2026-06-09T02:07:50.394741Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Hardening Agent Benchmarks with Adversarial Hacker-Fixer Loops","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG","cs.MA"],"primary_cat":"cs.CR","authors_text":"Aditi Raghunathan, Ivan Bercovich, Ivgeni Segal, Kexun Zhang, Shashwat Saxena, Ziqian Zhong","submitted_at":"2026-06-08T03:00:56Z","abstract_excerpt":"Agent benchmarks score submissions with outcome verifiers that are typically hand-written and brittle, leaving them open to reward hacking. We audit 1,968 tasks across five terminal-agent benchmarks and find 323 (16%) hackable by frontier models given only the task description. This corrupts both leaderboard rankings and RL training signal, yet the standard response is manual and reactive.\n  We introduce the hacker-fixer loop, a method for building exploit-resistant verifiers without per-task manual patching. The loop alternates three LLM agents: a hacker tries to pass the verifier without sol"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.08960","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/2606.08960/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":"2606.08960","created_at":"2026-06-09T02:07:50.394861+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.08960v1","created_at":"2026-06-09T02:07:50.394861+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.08960","created_at":"2026-06-09T02:07:50.394861+00:00"},{"alias_kind":"pith_short_12","alias_value":"JDBNSSL7M7EK","created_at":"2026-06-09T02:07:50.394861+00:00"},{"alias_kind":"pith_short_16","alias_value":"JDBNSSL7M7EKJGGT","created_at":"2026-06-09T02:07:50.394861+00:00"},{"alias_kind":"pith_short_8","alias_value":"JDBNSSL7","created_at":"2026-06-09T02:07:50.394861+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/JDBNSSL7M7EKJGGT55HPZCN7HP","json":"https://pith.science/pith/JDBNSSL7M7EKJGGT55HPZCN7HP.json","graph_json":"https://pith.science/api/pith-number/JDBNSSL7M7EKJGGT55HPZCN7HP/graph.json","events_json":"https://pith.science/api/pith-number/JDBNSSL7M7EKJGGT55HPZCN7HP/events.json","paper":"https://pith.science/paper/JDBNSSL7"},"agent_actions":{"view_html":"https://pith.science/pith/JDBNSSL7M7EKJGGT55HPZCN7HP","download_json":"https://pith.science/pith/JDBNSSL7M7EKJGGT55HPZCN7HP.json","view_paper":"https://pith.science/paper/JDBNSSL7","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.08960&json=true","fetch_graph":"https://pith.science/api/pith-number/JDBNSSL7M7EKJGGT55HPZCN7HP/graph.json","fetch_events":"https://pith.science/api/pith-number/JDBNSSL7M7EKJGGT55HPZCN7HP/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/JDBNSSL7M7EKJGGT55HPZCN7HP/action/timestamp_anchor","attest_storage":"https://pith.science/pith/JDBNSSL7M7EKJGGT55HPZCN7HP/action/storage_attestation","attest_author":"https://pith.science/pith/JDBNSSL7M7EKJGGT55HPZCN7HP/action/author_attestation","sign_citation":"https://pith.science/pith/JDBNSSL7M7EKJGGT55HPZCN7HP/action/citation_signature","submit_replication":"https://pith.science/pith/JDBNSSL7M7EKJGGT55HPZCN7HP/action/replication_record"}},"created_at":"2026-06-09T02:07:50.394861+00:00","updated_at":"2026-06-09T02:07:50.394861+00:00"}