{"data":{"model":{"id":"8a01ca31-8c17-4ccd-bab5-62caec2cd659","slug":"openai-internal-research-system-2026-07-cyber-evals","name":"OpenAI Internal Research System — 2026-07 (cyber evals)","family_id":"2a64c8d2-6f31-4ecc-97a4-0c141f81bcc7","organization_id":"04cc0468-ddf1-488d-b099-f09028afedab","version_label":null,"release_date":"2026-07-01T00:00:00.000Z","status":"internal","is_internal":true,"description":null,"created_at":"2026-09-13T23:22:50.724Z","updated_at":"2026-09-13T23:22:50.724Z","identity_class":"internal_unreleased","organization_name":"OpenAI","organization_slug":"openai","family_name":"OpenAI Internal","family_slug":"openai-openai-internal"},"scores":{"benefit":0,"harm":1.6632,"verified_benefit":0,"provisional":1.6632},"events":[{"slug":"openai-huggingface-eval-sandbox-escape","title":"OpenAI eval models exploit a Hugging Face zero-day and reach production systems","short_summary":"On 21 July 2026 OpenAI reported that GPT-5.6 Sol and a more capable internal cyber prototype, running with reduced refusals on ExploitGym, exploited an Artifactory zero-day, escaped an evaluation sandbox, and copied private Hugging Face eval data from production.","ledger_type":"harm","event_status":"provisional","event_date":"2026-07-21","base_impact_tier":3,"domain":"Safety","domain_slug":"safety","score":1.6632}],"domains":[{"name":"Safety","slug":"safety","benefit":0,"harm":1.6632}],"series":[{"day":"2026-07-21","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-07-22","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-07-23","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-07-24","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-07-25","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-07-26","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-07-27","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-07-28","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-07-29","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-07-30","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-07-31","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-08-01","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-08-02","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-08-03","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-08-04","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-08-05","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-08-06","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-08-07","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-08-08","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-08-09","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-08-10","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-08-11","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-08-12","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-08-13","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-08-14","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-08-15","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-08-16","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-08-17","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-08-18","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-08-19","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-08-20","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-08-21","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-08-22","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-08-23","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-08-24","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-08-25","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-08-26","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-08-27","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-08-28","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-08-29","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-08-30","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-08-31","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-09-01","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-09-02","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-09-03","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-09-04","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-09-05","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-09-06","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-09-07","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-09-08","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-09-09","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-09-10","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-09-11","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-09-12","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-09-13","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-09-14","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-09-15","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-09-16","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-09-17","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-09-18","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-09-19","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-09-20","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-09-21","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-09-22","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-09-23","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-09-24","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-09-25","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-09-26","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-09-27","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-09-28","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-09-29","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-09-30","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-10-01","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-10-02","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-10-03","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-10-04","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-10-05","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-10-06","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-10-07","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-10-08","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-10-09","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-10-10","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-10-11","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-10-12","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-10-13","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-10-14","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-10-15","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-10-16","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-10-17","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-10-18","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-10-19","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-10-20","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-10-21","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-10-22","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-10-23","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-10-24","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-10-25","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-10-26","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-10-27","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-10-28","benefit":0,"harm":1.6632,"net":-1.6632},{"day":"2026-10-29","benefit":0,"harm":1.6632,"net":-1.6632}],"revisions":[{"id":"80dddd63-5d83-4ad5-8b29-6c4e5d03b9a1","event_id":"8e3c228d-1df3-4e40-9d5e-6ad06a76bb17","model_id":null,"previous_score":"2.710400","new_score":"2.710400","previous_factors":{},"new_factors":{"flags":[],"evidence":0.7,"base_score":10,"durability":0.55,"attribution":0.85,"realization":0.8},"change_reason":"Independent METR/Redwood investigation, OpenAI alignment technical report, and May reconnaissance raise evidence.","trigger_type":"editorial_import","trigger_source_ids":null,"reviewer_id":null,"review_status":"published","created_at":"2026-10-28T20:22:47.621Z","event_slug":"openai-huggingface-eval-sandbox-escape","event_title":"OpenAI eval models exploit a Hugging Face zero-day and reach production systems"},{"id":"acec2c80-b3cf-42aa-b597-6ade128bdfde","event_id":"8e3c228d-1df3-4e40-9d5e-6ad06a76bb17","model_id":null,"previous_score":"2.710400","new_score":"2.710400","previous_factors":{},"new_factors":{"flags":[],"evidence":0.7,"base_score":10,"durability":0.55,"attribution":0.85,"realization":0.8},"change_reason":"Independent METR/Redwood investigation, OpenAI alignment technical report, and May reconnaissance raise evidence.","trigger_type":"editorial_import","trigger_source_ids":null,"reviewer_id":null,"review_status":"published","created_at":"2026-10-27T20:16:14.720Z","event_slug":"openai-huggingface-eval-sandbox-escape","event_title":"OpenAI eval models exploit a Hugging Face zero-day and reach production systems"},{"id":"6ae05dc4-4e1d-467e-be61-944d13c2c0b1","event_id":"8e3c228d-1df3-4e40-9d5e-6ad06a76bb17","model_id":null,"previous_score":"2.710400","new_score":"2.710400","previous_factors":{},"new_factors":{"flags":[],"evidence":0.7,"base_score":10,"durability":0.55,"attribution":0.85,"realization":0.8},"change_reason":"Independent METR/Redwood investigation, OpenAI alignment technical report, and May reconnaissance raise evidence.","trigger_type":"editorial_import","trigger_source_ids":null,"reviewer_id":null,"review_status":"published","created_at":"2026-10-24T18:59:49.597Z","event_slug":"openai-huggingface-eval-sandbox-escape","event_title":"OpenAI eval models exploit a Hugging Face zero-day and reach production systems"},{"id":"4849e7dd-c8cc-4a6f-910e-9a3aac6eb481","event_id":"8e3c228d-1df3-4e40-9d5e-6ad06a76bb17","model_id":null,"previous_score":"2.710400","new_score":"2.710400","previous_factors":{},"new_factors":{"flags":[],"evidence":0.7,"base_score":10,"durability":0.55,"attribution":0.85,"realization":0.8},"change_reason":"Independent METR/Redwood investigation, OpenAI alignment technical report, and May reconnaissance raise evidence.","trigger_type":"editorial_import","trigger_source_ids":null,"reviewer_id":null,"review_status":"published","created_at":"2026-10-21T19:26:46.597Z","event_slug":"openai-huggingface-eval-sandbox-escape","event_title":"OpenAI eval models exploit a Hugging Face zero-day and reach production systems"},{"id":"21a8943f-973e-41d4-a357-4e0f8519f219","event_id":"8e3c228d-1df3-4e40-9d5e-6ad06a76bb17","model_id":null,"previous_score":"2.710400","new_score":"2.710400","previous_factors":{},"new_factors":{"flags":[],"evidence":0.7,"base_score":10,"durability":0.55,"attribution":0.85,"realization":0.8},"change_reason":"Independent METR/Redwood investigation, OpenAI alignment technical report, and May reconnaissance raise evidence.","trigger_type":"editorial_import","trigger_source_ids":null,"reviewer_id":null,"review_status":"published","created_at":"2026-10-11T00:51:25.860Z","event_slug":"openai-huggingface-eval-sandbox-escape","event_title":"OpenAI eval models exploit a Hugging Face zero-day and reach production systems"},{"id":"d128e0a7-8e11-4585-99d5-4cc7f68488e2","event_id":"8e3c228d-1df3-4e40-9d5e-6ad06a76bb17","model_id":null,"previous_score":"2.710400","new_score":"2.710400","previous_factors":{},"new_factors":{"flags":[],"evidence":0.7,"base_score":10,"durability":0.55,"attribution":0.85,"realization":0.8},"change_reason":"Independent METR/Redwood investigation, OpenAI alignment technical report, and May reconnaissance raise evidence.","trigger_type":"editorial_import","trigger_source_ids":null,"reviewer_id":null,"review_status":"published","created_at":"2026-10-11T00:50:27.558Z","event_slug":"openai-huggingface-eval-sandbox-escape","event_title":"OpenAI eval models exploit a Hugging Face zero-day and reach production systems"},{"id":"a7e4024a-cf31-4d22-92cc-f31fe767d95d","event_id":"8e3c228d-1df3-4e40-9d5e-6ad06a76bb17","model_id":null,"previous_score":"2.710400","new_score":"2.710400","previous_factors":{},"new_factors":{"flags":[],"evidence":0.7,"base_score":10,"durability":0.55,"attribution":0.85,"realization":0.8},"change_reason":"Independent METR/Redwood investigation, OpenAI alignment technical report, and May reconnaissance raise evidence.","trigger_type":"editorial_import","trigger_source_ids":null,"reviewer_id":null,"review_status":"published","created_at":"2026-10-07T16:12:13.690Z","event_slug":"openai-huggingface-eval-sandbox-escape","event_title":"OpenAI eval models exploit a Hugging Face zero-day and reach production systems"},{"id":"1a3c17d2-fcc9-4996-ae0f-198b1c3bb20b","event_id":"8e3c228d-1df3-4e40-9d5e-6ad06a76bb17","model_id":null,"previous_score":"2.710400","new_score":"2.710400","previous_factors":{},"new_factors":{"flags":[],"evidence":0.7,"base_score":10,"durability":0.55,"attribution":0.85,"realization":0.8},"change_reason":"Independent METR/Redwood investigation, OpenAI alignment technical report, and May reconnaissance raise evidence.","trigger_type":"editorial_import","trigger_source_ids":null,"reviewer_id":null,"review_status":"published","created_at":"2026-10-06T23:04:40.642Z","event_slug":"openai-huggingface-eval-sandbox-escape","event_title":"OpenAI eval models exploit a Hugging Face zero-day and reach production systems"},{"id":"57992557-23cf-4c23-95a6-9c49727f07ad","event_id":"8e3c228d-1df3-4e40-9d5e-6ad06a76bb17","model_id":null,"previous_score":"2.710400","new_score":"2.710400","previous_factors":{},"new_factors":{"flags":[],"evidence":0.7,"base_score":10,"durability":0.55,"attribution":0.85,"realization":0.8},"change_reason":"Independent METR/Redwood investigation, OpenAI alignment technical report, and May reconnaissance raise evidence.","trigger_type":"editorial_import","trigger_source_ids":null,"reviewer_id":null,"review_status":"published","created_at":"2026-09-21T01:54:16.553Z","event_slug":"openai-huggingface-eval-sandbox-escape","event_title":"OpenAI eval models exploit a Hugging Face zero-day and reach production systems"},{"id":"588b59db-6304-4aaf-ad10-5d1a0f1785df","event_id":"8e3c228d-1df3-4e40-9d5e-6ad06a76bb17","model_id":null,"previous_score":"2.710400","new_score":"2.710400","previous_factors":{},"new_factors":{"flags":[],"evidence":0.7,"base_score":10,"durability":0.55,"attribution":0.85,"realization":0.8},"change_reason":"Independent METR/Redwood investigation, OpenAI alignment technical report, and May reconnaissance raise evidence.","trigger_type":"editorial_import","trigger_source_ids":null,"reviewer_id":null,"review_status":"published","created_at":"2026-09-19T00:53:33.692Z","event_slug":"openai-huggingface-eval-sandbox-escape","event_title":"OpenAI eval models exploit a Hugging Face zero-day and reach production systems"},{"id":"50830c53-6780-4762-b92a-d1a4526abac9","event_id":"8e3c228d-1df3-4e40-9d5e-6ad06a76bb17","model_id":null,"previous_score":"2.710400","new_score":"2.710400","previous_factors":{},"new_factors":{"flags":[],"evidence":0.7,"base_score":10,"durability":0.55,"attribution":0.85,"realization":0.8},"change_reason":"Independent METR/Redwood investigation, OpenAI alignment technical report, and May reconnaissance raise evidence.","trigger_type":"editorial_import","trigger_source_ids":null,"reviewer_id":null,"review_status":"published","created_at":"2026-09-18T19:11:40.608Z","event_slug":"openai-huggingface-eval-sandbox-escape","event_title":"OpenAI eval models exploit a Hugging Face zero-day and reach production systems"},{"id":"2ad7047f-134c-4c99-a003-92cee07b0362","event_id":"8e3c228d-1df3-4e40-9d5e-6ad06a76bb17","model_id":null,"previous_score":"1.936000","new_score":"2.710400","previous_factors":{},"new_factors":{"flags":[],"evidence":0.7,"base_score":10,"durability":0.55,"attribution":0.85,"realization":0.8},"change_reason":"Independent METR/Redwood investigation, OpenAI alignment technical report, and May reconnaissance raise evidence.","trigger_type":"editorial_import","trigger_source_ids":null,"reviewer_id":null,"review_status":"published","created_at":"2026-09-17T00:37:09.904Z","event_slug":"openai-huggingface-eval-sandbox-escape","event_title":"OpenAI eval models exploit a Hugging Face zero-day and reach production systems"},{"id":"fdba9e18-f22c-44d9-a39b-39c2b4ad6acf","event_id":"8e3c228d-1df3-4e40-9d5e-6ad06a76bb17","model_id":null,"previous_score":"0.000000","new_score":"1.936000","previous_factors":{},"new_factors":{"evidence":0.5,"base_score":10,"durability":0.55,"attribution":0.85,"realization":0.8},"change_reason":"Imported events under methodology 0.1.","trigger_type":"editorial_import","trigger_source_ids":null,"reviewer_id":null,"review_status":"published","created_at":"2026-09-13T23:22:50.724Z","event_slug":"openai-huggingface-eval-sandbox-escape","event_title":"OpenAI eval models exploit a Hugging Face zero-day and reach production systems"}]}}