{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:FEGJT5FVVGFUWCKNQDJTFT4JDY","short_pith_number":"pith:FEGJT5FV","canonical_record":{"source":{"id":"2607.26115","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CR","submitted_at":"2026-07-28T16:03:39Z","cross_cats_sorted":["cs.AI","cs.CL","cs.LG"],"title_canon_sha256":"08ddba0c2fa2c32e181a17c520fcb1f84474430372d3bfe61c86c0c2992f8a2a","abstract_canon_sha256":"2a96eff108a469963959859950d26dd07ea43ee84dc70809d158431512c6025a"},"schema_version":"1.0"},"canonical_sha256":"290c99f4b5a98b4b094d80d332cf891e16e7d5be9b911a65e1272b385c17af64","source":{"kind":"arxiv","id":"2607.26115","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2607.26115","created_at":"2026-07-30T00:08:23Z"},{"alias_kind":"arxiv_version","alias_value":"2607.26115v1","created_at":"2026-07-30T00:08:23Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2607.26115","created_at":"2026-07-30T00:08:23Z"},{"alias_kind":"pith_short_12","alias_value":"FEGJT5FVVGFU","created_at":"2026-07-30T00:08:23Z"},{"alias_kind":"pith_short_16","alias_value":"FEGJT5FVVGFUWCKN","created_at":"2026-07-30T00:08:23Z"},{"alias_kind":"pith_short_8","alias_value":"FEGJT5FV","created_at":"2026-07-30T00:08:23Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:FEGJT5FVVGFUWCKNQDJTFT4JDY","target":"record","payload":{"canonical_record":{"source":{"id":"2607.26115","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CR","submitted_at":"2026-07-28T16:03:39Z","cross_cats_sorted":["cs.AI","cs.CL","cs.LG"],"title_canon_sha256":"08ddba0c2fa2c32e181a17c520fcb1f84474430372d3bfe61c86c0c2992f8a2a","abstract_canon_sha256":"2a96eff108a469963959859950d26dd07ea43ee84dc70809d158431512c6025a"},"schema_version":"1.0"},"canonical_sha256":"290c99f4b5a98b4b094d80d332cf891e16e7d5be9b911a65e1272b385c17af64","receipt":{"kind":"pith_receipt","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"290c99f4b5a98b4b094d80d332cf891e16e7d5be9b911a65e1272b385c17af64","last_reissued_at":"2026-07-30T00:08:23.145125Z","signature_status":"unsigned_v0","first_computed_at":"2026-07-30T00:08:23.145125Z"},"source_kind":"arxiv","source_id":"2607.26115","source_version":1,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-30T00:08:23Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"+4qmvixvoZuHMluKUDPKUNGAUbjE2rRlyt/rjV/7dZr5TfBbLU72U8ZE90A2AfocrSKRcrFWzUvB67kAA4yxDg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-04T02:02:10.483729Z"},"content_sha256":"0a2b5e6a3abad7cd116d4819a3c6826891ed17c224be5a412740926df0e2fe8e","schema_version":"1.0","event_id":"sha256:0a2b5e6a3abad7cd116d4819a3c6826891ed17c224be5a412740926df0e2fe8e"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:FEGJT5FVVGFUWCKNQDJTFT4JDY","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"GPT-Red: Automated Red Teaming via Self-Play at Scale","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CL","cs.LG"],"primary_cat":"cs.CR","authors_text":"Aiden Low, Christopher A. Choquette-Choo, Christopher Wolff, Chuan Guo, Dylan Hunn, Eric Wallace, Juan Felipe Cer\\'on Uribe, Kai Chen, Kaiwen Wang, Kai Xiao, Milad Nasr, Nikhil Kandpal, Sam Toyer, Sicheng Zhu, Stephanie Lin, Xiangyu Qi, Yuxin Wen, Zizhao Wang","submitted_at":"2026-07-28T16:03:39Z","abstract_excerpt":"We introduce \\textbf{GPT-Red}, an automated red-teaming agent that is trained to discover novel prompt injection attacks against frontier LLMs. The goal of this model is to evaluate and improve the robustness of our production systems. To this end, we use it to adversarially train GPT-5.6, our most robust model to prompt injections to date. To create GPT-Red, we design a scalable self-play algorithm where the model is tasked with attacking a diverse population of simultaneously-trained defender agents. We train the model on realistic red-teaming environments using compute on the same scale as "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2607.26115","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/2607.26115/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"},"verdict_id":null},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-30T00:08:23Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"JpEIcP46rKnsFC2/nAnHJ+/WqBnBdFstzA2ypcTIxYywMXW6jcwxtDCWrBU1eOmf1BAyyzWzZpcFG7HdVW+mCw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-04T02:02:10.484245Z"},"content_sha256":"62f1711f9a389259727b48eef4bfcdf91869553f50032ee7ba15ab117b63945f","schema_version":"1.0","event_id":"sha256:62f1711f9a389259727b48eef4bfcdf91869553f50032ee7ba15ab117b63945f"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/FEGJT5FVVGFUWCKNQDJTFT4JDY/bundle.json","state_url":"https://pith.science/pith/FEGJT5FVVGFUWCKNQDJTFT4JDY/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/FEGJT5FVVGFUWCKNQDJTFT4JDY/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-08-04T02:02:10Z","links":{"resolver":"https://pith.science/pith/FEGJT5FVVGFUWCKNQDJTFT4JDY","bundle":"https://pith.science/pith/FEGJT5FVVGFUWCKNQDJTFT4JDY/bundle.json","state":"https://pith.science/pith/FEGJT5FVVGFUWCKNQDJTFT4JDY/state.json","well_known_bundle":"https://pith.science/.well-known/pith/FEGJT5FVVGFUWCKNQDJTFT4JDY/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:FEGJT5FVVGFUWCKNQDJTFT4JDY","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"2a96eff108a469963959859950d26dd07ea43ee84dc70809d158431512c6025a","cross_cats_sorted":["cs.AI","cs.CL","cs.LG"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CR","submitted_at":"2026-07-28T16:03:39Z","title_canon_sha256":"08ddba0c2fa2c32e181a17c520fcb1f84474430372d3bfe61c86c0c2992f8a2a"},"schema_version":"1.0","source":{"id":"2607.26115","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2607.26115","created_at":"2026-07-30T00:08:23Z"},{"alias_kind":"arxiv_version","alias_value":"2607.26115v1","created_at":"2026-07-30T00:08:23Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2607.26115","created_at":"2026-07-30T00:08:23Z"},{"alias_kind":"pith_short_12","alias_value":"FEGJT5FVVGFU","created_at":"2026-07-30T00:08:23Z"},{"alias_kind":"pith_short_16","alias_value":"FEGJT5FVVGFUWCKN","created_at":"2026-07-30T00:08:23Z"},{"alias_kind":"pith_short_8","alias_value":"FEGJT5FV","created_at":"2026-07-30T00:08:23Z"}],"graph_snapshots":[{"event_id":"sha256:62f1711f9a389259727b48eef4bfcdf91869553f50032ee7ba15ab117b63945f","target":"graph","created_at":"2026-07-30T00:08:23Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2607.26115/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"We introduce \\textbf{GPT-Red}, an automated red-teaming agent that is trained to discover novel prompt injection attacks against frontier LLMs. The goal of this model is to evaluate and improve the robustness of our production systems. To this end, we use it to adversarially train GPT-5.6, our most robust model to prompt injections to date. To create GPT-Red, we design a scalable self-play algorithm where the model is tasked with attacking a diverse population of simultaneously-trained defender agents. We train the model on realistic red-teaming environments using compute on the same scale as ","authors_text":"Aiden Low, Christopher A. Choquette-Choo, Christopher Wolff, Chuan Guo, Dylan Hunn, Eric Wallace, Juan Felipe Cer\\'on Uribe, Kai Chen, Kaiwen Wang, Kai Xiao, Milad Nasr, Nikhil Kandpal, Sam Toyer, Sicheng Zhu, Stephanie Lin, Xiangyu Qi, Yuxin Wen, Zizhao Wang","cross_cats":["cs.AI","cs.CL","cs.LG"],"headline":"","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CR","submitted_at":"2026-07-28T16:03:39Z","title":"GPT-Red: Automated Red Teaming via Self-Play at Scale"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2607.26115","kind":"arxiv","version":1},"verdict":{"created_at":null,"id":null,"model_set":{},"one_line_summary":"","pipeline_version":null,"pith_extraction_headline":"","strongest_claim":"","weakest_assumption":""}},"verdict_id":null}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:0a2b5e6a3abad7cd116d4819a3c6826891ed17c224be5a412740926df0e2fe8e","target":"record","created_at":"2026-07-30T00:08:23Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"2a96eff108a469963959859950d26dd07ea43ee84dc70809d158431512c6025a","cross_cats_sorted":["cs.AI","cs.CL","cs.LG"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CR","submitted_at":"2026-07-28T16:03:39Z","title_canon_sha256":"08ddba0c2fa2c32e181a17c520fcb1f84474430372d3bfe61c86c0c2992f8a2a"},"schema_version":"1.0","source":{"id":"2607.26115","kind":"arxiv","version":1}},"canonical_sha256":"290c99f4b5a98b4b094d80d332cf891e16e7d5be9b911a65e1272b385c17af64","receipt":{"builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"290c99f4b5a98b4b094d80d332cf891e16e7d5be9b911a65e1272b385c17af64","first_computed_at":"2026-07-30T00:08:23.145125Z","kind":"pith_receipt","last_reissued_at":"2026-07-30T00:08:23.145125Z","receipt_version":"0.3","signature_status":"unsigned_v0"},"source_id":"2607.26115","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:0a2b5e6a3abad7cd116d4819a3c6826891ed17c224be5a412740926df0e2fe8e","sha256:62f1711f9a389259727b48eef4bfcdf91869553f50032ee7ba15ab117b63945f"],"state_sha256":"3af41a30c3ba221d4a79c7909780eeb3651ab265a4a53efcb729d9f24cdb506f"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"HdG7RigffB6JexKHdq9Z9uySerEBqXp6ndygVLWEAq5y2NkkeRMqnxu6qqLbtDDsm5Zo2TW76/2OnBKPkOs3AA==","signed_message":"bundle_sha256_bytes","signed_at":"2026-08-04T02:02:10.489136Z","bundle_sha256":"89e9110a6dc1dcfdc78a33a65dd3d1455c7691225f9558be3a4cf6d0f36901e3"}}