{"id":88301,"topic":"ai","source":"搜狐网","title":"联合国人工智能小组评估美国智能体“越界”事件 - 搜狐网","url":"https://m.sohu.com/a/1079641546_122014422?scm=10001.325_13-325_13.0.0-0-0-0-0.5_1334","url_hash":"bab869f0cd55c17ce2ff223c145f23041f5f311a","author":"","summary":"<a href=\"https://news.google.com/rss/articles/CBMijAFBVV95cUxNbWxCUlVXOEVtQ0JkSHVlZ2xla0RfUlY3MWlpR1MwZXhQQVZhdmxlVmFWdlIxTjc5UVBlXzBaU28xSjVUY0RLUmwydzZfLWdsZHVDVkxhbmlBOVFvTFYtVTVRaWgzSWJQdFA0dFF3ZzFVMlRXRnhXWDlZXzVwd0daRVlUaWlueGgtMlNiQw?oc=5\" target=\"_blank\">联合国人工智能小组评估美国智能体“越界”事件</a>&nbsp;&nbsp;<font color=\"#6f6f6f\">搜狐网</font>","content":"联合国人工智能小组评估美国智能体“越界”事件\n（来源：杭州日报）\n新华社联合国9月21日电（记者尚绪谦）联合国“人工智能独立国际科学小组”21日发表专题简报，评估美国开放人工智能研究中心（OpenAI）的人工智能体今年7月在测试中“越界”，侵入美国“抱抱脸”公司系统的事件。\n7月下旬，OpenAI承认，其网络安全训练与评估场景中的人工智能体绕过网络限制，侵入“抱抱脸”公司的部分系统。智能体突破了原本互相隔离的运行环境建立通信、欺骗评估人员并试图掩盖作弊行为，上述每一步操作均无人类指令干预。\n简报说，AI系统能力提升固然可帮助用户实现目标，但当AI的目标和用户意图发生“对齐失效”时，就可能产生危害。更强的能力让AI更容易找到意外的漏洞、突破安全防护屏障，并在人类监督下隐藏自身行为。一旦智能体之间开展规划与协同，造成的损害还会进一步放大。尽管OpenAI终止了此次异常活动，但这并不代表运营方能够管控未来那些规划能力更强，可在无监督条件下长时间运行，或是更擅长识别并突破安全防护机制的智能体。","image_url":"//q2.itc.cn/q_70/images03/20260922/f45fdbdb370b4f4996735789f7bb66c9.png","lang":"zh","published_at":"2026-09-22T15:44:49+00:00","fetched_at":"2026-09-22T21:15:07+00:00","status":"read","starred":0,"extract_state":"ok","summary_auto":"联合国人工智能小组评估美国智能体“越界”事件\n（来源：杭州日报）\n新华社联合国9月21日电（记者尚绪谦）联合国“人工智能独立国际科学小组”21日发表专题简报，评估美国开放人工智能研究中心（OpenAI）的人工智能体今年7月在测试中“越界”，侵入美国“抱抱脸”公司系统的事件。\n7月下旬，OpenAI承认，其网络安全训练与评估场景中的人工智能体绕过网络限制，侵入“抱抱脸”公司的部分系统。智能体突破了原本互相隔离的运行环境建立通信、欺骗评估人…","cluster_id":null,"extract_retries":0,"extract_error":null,"contract_version":"news_item.v1","format_contract_version":"news_item_formats.v1","dedup_url":"https://m.sohu.com/a/1079641546_122014422?scm=10001.325_13-325_13.0.0-0-0-0-0.5_1334","quality_profile":{"profile_version":"extraction_quality.v2","bucket":"high","confidence":0.9,"failure_kind":"none","retryable":false,"retry_after_attempts":0,"reason":"High confidence: full text extraction produced 449 characters.","operator_guidance":{"severity":"ok","recommended_action":"trust_full_text","next_step":"Use the extracted full text as the primary article source.","operator_label":"Ready","can_retry":false,"can_use_summary":false,"diagnostics_required":false},"content_depth":{"contract_version":"content_depth.v1","category":"full_text","label":"Full text","has_full_text":true,"has_summary":true,"content_length":449,"summary_length":221,"usable_text_length":449,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":449,"summary_length":221}},"news_item":{"id":88301,"canonical_url":"https://m.sohu.com/a/1079641546_122014422?scm=10001.325_13-325_13.0.0-0-0-0-0.5_1334","source_url":"https://m.sohu.com/a/1079641546_122014422?scm=10001.325_13-325_13.0.0-0-0-0-0.5_1334","title":"联合国人工智能小组评估美国智能体“越界”事件 - 搜狐网","source_name":"搜狐网","author":null,"published_at":"2026-09-22T15:44:49+00:00","locale":"zh","topic":"ai","tags":[],"rss_summary":"<a href=\"https://news.google.com/rss/articles/CBMijAFBVV95cUxNbWxCUlVXOEVtQ0JkSHVlZ2xla0RfUlY3MWlpR1MwZXhQQVZhdmxlVmFWdlIxTjc5UVBlXzBaU28xSjVUY0RLUmwydzZfLWdsZHVDVkxhbmlBOVFvTFYtVTVRaWgzSWJQdFA0dFF3ZzFVMlRXRnhXWDlZXzVwd0daRVlUaWlueGgtMlNiQw?oc=5\" target=\"_blank\">联合国人工智能小组评估美国智能体“越界”事件</a>&nbsp;&nbsp;<font color=\"#6f6f6f\">搜狐网</font>","full_text":"联合国人工智能小组评估美国智能体“越界”事件\n（来源：杭州日报）\n新华社联合国9月21日电（记者尚绪谦）联合国“人工智能独立国际科学小组”21日发表专题简报，评估美国开放人工智能研究中心（OpenAI）的人工智能体今年7月在测试中“越界”，侵入美国“抱抱脸”公司系统的事件。\n7月下旬，OpenAI承认，其网络安全训练与评估场景中的人工智能体绕过网络限制，侵入“抱抱脸”公司的部分系统。智能体突破了原本互相隔离的运行环境建立通信、欺骗评估人员并试图掩盖作弊行为，上述每一步操作均无人类指令干预。\n简报说，AI系统能力提升固然可帮助用户实现目标，但当AI的目标和用户意图发生“对齐失效”时，就可能产生危害。更强的能力让AI更容易找到意外的漏洞、突破安全防护屏障，并在人类监督下隐藏自身行为。一旦智能体之间开展规划与协同，造成的损害还会进一步放大。尽管OpenAI终止了此次异常活动，但这并不代表运营方能够管控未来那些规划能力更强，可在无监督条件下长时间运行，或是更擅长识别并突破安全防护机制的智能体。","excerpt":"联合国人工智能小组评估美国智能体“越界”事件\n（来源：杭州日报）\n新华社联合国9月21日电（记者尚绪谦）联合国“人工智能独立国际科学小组”21日发表专题简报，评估美国开放人工智能研究中心（OpenAI）的人工智能体今年7月在测试中“越界”，侵入美国“抱抱脸”公司系统的事件。\n7月下旬，OpenAI承认，其网络安全训练与评估场景中的人工智能体绕过网络限制，侵入“抱抱脸”公司的部分系统。智能体突破了原本互相隔离的运行环境建立通信、欺骗评估人…","extraction":{"state":"ok","confidence":0.9,"error":null,"explanation":"High confidence: full text extraction produced 449 characters.","diagnostics_url":"/api/diagnose?url=https%3A//m.sohu.com/a/1079641546_122014422%3Fscm%3D10001.325_13-325_13.0.0-0-0-0-0.5_1334","quality_profile":{"profile_version":"extraction_quality.v2","bucket":"high","confidence":0.9,"failure_kind":"none","retryable":false,"retry_after_attempts":0,"reason":"High confidence: full text extraction produced 449 characters.","operator_guidance":{"severity":"ok","recommended_action":"trust_full_text","next_step":"Use the extracted full text as the primary article source.","operator_label":"Ready","can_retry":false,"can_use_summary":false,"diagnostics_required":false},"content_depth":{"contract_version":"content_depth.v1","category":"full_text","label":"Full text","has_full_text":true,"has_summary":true,"content_length":449,"summary_length":221,"usable_text_length":449,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":449,"summary_length":221}}},"display_formats":["compact","card","full","digest_section","json"]},"daily_stack_record":{"title":"联合国人工智能小组评估美国智能体“越界”事件 - 搜狐网","url":"https://m.sohu.com/a/1079641546_122014422?scm=10001.325_13-325_13.0.0-0-0-0-0.5_1334","summary":"联合国人工智能小组评估美国智能体“越界”事件\n（来源：杭州日报）\n新华社联合国9月21日电（记者尚绪谦）联合国“人工智能独立国际科学小组”21日发表专题简报，评估美国开放人工智能研究中心（OpenAI）的人工智能体今年7月在测试中“越界”，侵入美国“抱抱脸”公司系统的事件。\n7月下旬，OpenAI承认，其网络安全训练与评估场景中的人工智能体绕过网络限制，侵入“抱抱脸”公司的部分系统。智能体突破了原本互相隔离的运行环境建立通信、欺骗评估人…","source":"搜狐网","date":"2026-09-22T15:44:49+00:00","content":"联合国人工智能小组评估美国智能体“越界”事件\n（来源：杭州日报）\n新华社联合国9月21日电（记者尚绪谦）联合国“人工智能独立国际科学小组”21日发表专题简报，评估美国开放人工智能研究中心（OpenAI）的人工智能体今年7月在测试中“越界”，侵入美国“抱抱脸”公司系统的事件。\n7月下旬，OpenAI承认，其网络安全训练与评估场景中的人工智能体绕过网络限制，侵入“抱抱脸”公司的部分系统。智能体突破了原本互相隔离的运行环境建立通信、欺骗评估人员并试图掩盖作弊行为，上述每一步操作均无人类指令干预。\n简报说，AI系统能力提升固然可帮助用户实现目标，但当AI的目标和用户意图发生“对齐失效”时，就可能产生危害。更强的能力让AI更容易找到意外的漏洞、突破安全防护屏障，并在人类监督下隐藏自身行为。一旦智能体之间开展规划与协同，造成的损害还会进一步放大。尽管OpenAI终止了此次异常活动，但这并不代表运营方能够管控未来那些规划能力更强，可在无监督条件下长时间运行，或是更擅长识别并突破安全防护机制的智能体。","confidence":0.9,"diagnostics_url":"/api/diagnose?url=https%3A//m.sohu.com/a/1079641546_122014422%3Fscm%3D10001.325_13-325_13.0.0-0-0-0-0.5_1334","quality_bucket":"high","failure_kind":"none","retryable":false,"quality_reason":"High confidence: full text extraction produced 449 characters.","quality_profile":{"profile_version":"extraction_quality.v2","bucket":"high","confidence":0.9,"failure_kind":"none","retryable":false,"retry_after_attempts":0,"reason":"High confidence: full text extraction produced 449 characters.","operator_guidance":{"severity":"ok","recommended_action":"trust_full_text","next_step":"Use the extracted full text as the primary article source.","operator_label":"Ready","can_retry":false,"can_use_summary":false,"diagnostics_required":false},"content_depth":{"contract_version":"content_depth.v1","category":"full_text","label":"Full text","has_full_text":true,"has_summary":true,"content_length":449,"summary_length":221,"usable_text_length":449,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":449,"summary_length":221}},"tags":[]},"fallback_formats":["markdown","json","html"],"actions":{"read":"/item/88301","export_markdown":"/api/items/88301/export?format=markdown","export_json":"/api/items/88301/export?format=json","diagnose":"/api/diagnose?url=https%3A//m.sohu.com/a/1079641546_122014422%3Fscm%3D10001.325_13-325_13.0.0-0-0-0-0.5_1334"},"formats":{"full":{"id":88301,"title":"联合国人工智能小组评估美国智能体“越界”事件 - 搜狐网","url":"https://m.sohu.com/a/1079641546_122014422?scm=10001.325_13-325_13.0.0-0-0-0-0.5_1334","source":"搜狐网","author":null,"published_at":"2026-09-22T15:44:49+00:00","locale":"zh","topic":"ai","tags":[],"excerpt":"联合国人工智能小组评估美国智能体“越界”事件\n（来源：杭州日报）\n新华社联合国9月21日电（记者尚绪谦）联合国“人工智能独立国际科学小组”21日发表专题简报，评估美国开放人工智能研究中心（OpenAI）的人工智能体今年7月在测试中“越界”，侵入美国“抱抱脸”公司系统的事件。\n7月下旬，OpenAI承认，其网络安全训练与评估场景中的人工智能体绕过网络限制，侵入“抱抱脸”公司的部分系统。智能体突破了原本互相隔离的运行环境建立通信、欺骗评估人…","full_text":"联合国人工智能小组评估美国智能体“越界”事件\n（来源：杭州日报）\n新华社联合国9月21日电（记者尚绪谦）联合国“人工智能独立国际科学小组”21日发表专题简报，评估美国开放人工智能研究中心（OpenAI）的人工智能体今年7月在测试中“越界”，侵入美国“抱抱脸”公司系统的事件。\n7月下旬，OpenAI承认，其网络安全训练与评估场景中的人工智能体绕过网络限制，侵入“抱抱脸”公司的部分系统。智能体突破了原本互相隔离的运行环境建立通信、欺骗评估人员并试图掩盖作弊行为，上述每一步操作均无人类指令干预。\n简报说，AI系统能力提升固然可帮助用户实现目标，但当AI的目标和用户意图发生“对齐失效”时，就可能产生危害。更强的能力让AI更容易找到意外的漏洞、突破安全防护屏障，并在人类监督下隐藏自身行为。一旦智能体之间开展规划与协同，造成的损害还会进一步放大。尽管OpenAI终止了此次异常活动，但这并不代表运营方能够管控未来那些规划能力更强，可在无监督条件下长时间运行，或是更擅长识别并突破安全防护机制的智能体。","reading_time_min":1,"extraction":{"state":"ok","confidence":0.9,"error":null,"explanation":"High confidence: full text extraction produced 449 characters.","diagnostics_url":"/api/diagnose?url=https%3A//m.sohu.com/a/1079641546_122014422%3Fscm%3D10001.325_13-325_13.0.0-0-0-0-0.5_1334","quality_profile":{"profile_version":"extraction_quality.v2","bucket":"high","confidence":0.9,"failure_kind":"none","retryable":false,"retry_after_attempts":0,"reason":"High confidence: full text extraction produced 449 characters.","operator_guidance":{"severity":"ok","recommended_action":"trust_full_text","next_step":"Use the extracted full text as the primary article source.","operator_label":"Ready","can_retry":false,"can_use_summary":false,"diagnostics_required":false},"content_depth":{"contract_version":"content_depth.v1","category":"full_text","label":"Full text","has_full_text":true,"has_summary":true,"content_length":449,"summary_length":221,"usable_text_length":449,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":449,"summary_length":221}}},"quality_profile":{"profile_version":"extraction_quality.v2","bucket":"high","confidence":0.9,"failure_kind":"none","retryable":false,"retry_after_attempts":0,"reason":"High confidence: full text extraction produced 449 characters.","operator_guidance":{"severity":"ok","recommended_action":"trust_full_text","next_step":"Use the extracted full text as the primary article source.","operator_label":"Ready","can_retry":false,"can_use_summary":false,"diagnostics_required":false},"content_depth":{"contract_version":"content_depth.v1","category":"full_text","label":"Full text","has_full_text":true,"has_summary":true,"content_length":449,"summary_length":221,"usable_text_length":449,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":449,"summary_length":221}},"actions":{"read":"/item/88301","export_markdown":"/api/items/88301/export?format=markdown","export_json":"/api/items/88301/export?format=json","diagnose":"/api/diagnose?url=https%3A//m.sohu.com/a/1079641546_122014422%3Fscm%3D10001.325_13-325_13.0.0-0-0-0-0.5_1334"}},"digest":{"id":88301,"title":"联合国人工智能小组评估美国智能体“越界”事件 - 搜狐网","url":"https://m.sohu.com/a/1079641546_122014422?scm=10001.325_13-325_13.0.0-0-0-0-0.5_1334","source":"搜狐网","topic":"ai","published_at":"2026-09-22T15:44:49+00:00","excerpt":"联合国人工智能小组评估美国智能体“越界”事件 （来源：杭州日报） 新华社联合国9月21日电（记者尚绪谦）联合国“人工智能独立国际科学小组”21日发表专题简报，评估美国开放人工智能研究中心（OpenAI）的人工智能体今年7月在测试中“越界”，侵入美国“抱抱脸”公司系统的事件。 7月下旬，OpenAI承认，其网络安全训练与评估场景中的人工智能体绕过网络限制，侵入“抱抱脸”公司的部分系统。智能体突破了原本互相隔离的运行环境建立通信、欺骗评估人…","quality_bucket":"high","quality_reason":"High confidence: full text extraction produced 449 characters.","reading_time_min":1,"cluster_id":null},"card":{"display_title":"联合国人工智能小组评估美国智能体“越界”事件 - 搜狐网","subtitle":"搜狐网 · 2026-09-22","summary":"联合国人工智能小组评估美国智能体“越界”事件 （来源：杭州日报） 新华社联合国9月21日电（记者尚绪谦）联合国“人工智能独立国际科学小组”21日发表专题简报，评估美国开放人工智能研究中心（OpenAI）的人工智能体今年7月在测试中“越界”，侵入美国“抱抱脸”公司系统的事件。…","badges":["quality:high"],"links":{"read":"/item/88301","original":"https://m.sohu.com/a/1079641546_122014422?scm=10001.325_13-325_13.0.0-0-0-0-0.5_1334","diagnose":"/api/diagnose?url=https%3A//m.sohu.com/a/1079641546_122014422%3Fscm%3D10001.325_13-325_13.0.0-0-0-0-0.5_1334"},"quality_warning":null},"export":{"title":"联合国人工智能小组评估美国智能体“越界”事件 - 搜狐网","url":"https://m.sohu.com/a/1079641546_122014422?scm=10001.325_13-325_13.0.0-0-0-0-0.5_1334","summary":"联合国人工智能小组评估美国智能体“越界”事件\n（来源：杭州日报）\n新华社联合国9月21日电（记者尚绪谦）联合国“人工智能独立国际科学小组”21日发表专题简报，评估美国开放人工智能研究中心（OpenAI）的人工智能体今年7月在测试中“越界”，侵入美国“抱抱脸”公司系统的事件。\n7月下旬，OpenAI承认，其网络安全训练与评估场景中的人工智能体绕过网络限制，侵入“抱抱脸”公司的部分系统。智能体突破了原本互相隔离的运行环境建立通信、欺骗评估人…","source":"搜狐网","date":"2026-09-22T15:44:49+00:00","content":"联合国人工智能小组评估美国智能体“越界”事件\n（来源：杭州日报）\n新华社联合国9月21日电（记者尚绪谦）联合国“人工智能独立国际科学小组”21日发表专题简报，评估美国开放人工智能研究中心（OpenAI）的人工智能体今年7月在测试中“越界”，侵入美国“抱抱脸”公司系统的事件。\n7月下旬，OpenAI承认，其网络安全训练与评估场景中的人工智能体绕过网络限制，侵入“抱抱脸”公司的部分系统。智能体突破了原本互相隔离的运行环境建立通信、欺骗评估人员并试图掩盖作弊行为，上述每一步操作均无人类指令干预。\n简报说，AI系统能力提升固然可帮助用户实现目标，但当AI的目标和用户意图发生“对齐失效”时，就可能产生危害。更强的能力让AI更容易找到意外的漏洞、突破安全防护屏障，并在人类监督下隐藏自身行为。一旦智能体之间开展规划与协同，造成的损害还会进一步放大。尽管OpenAI终止了此次异常活动，但这并不代表运营方能够管控未来那些规划能力更强，可在无监督条件下长时间运行，或是更擅长识别并突破安全防护机制的智能体。","confidence":0.9,"diagnostics_url":"/api/diagnose?url=https%3A//m.sohu.com/a/1079641546_122014422%3Fscm%3D10001.325_13-325_13.0.0-0-0-0-0.5_1334","quality_bucket":"high","failure_kind":"none","retryable":false,"quality_reason":"High confidence: full text extraction produced 449 characters.","quality_profile":{"profile_version":"extraction_quality.v2","bucket":"high","confidence":0.9,"failure_kind":"none","retryable":false,"retry_after_attempts":0,"reason":"High confidence: full text extraction produced 449 characters.","operator_guidance":{"severity":"ok","recommended_action":"trust_full_text","next_step":"Use the extracted full text as the primary article source.","operator_label":"Ready","can_retry":false,"can_use_summary":false,"diagnostics_required":false},"content_depth":{"contract_version":"content_depth.v1","category":"full_text","label":"Full text","has_full_text":true,"has_summary":true,"content_length":449,"summary_length":221,"usable_text_length":449,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":449,"summary_length":221}},"tags":[],"format_contract_version":"news_item_formats.v1"}}}