{"id":85694,"topic":"ai","source":"环球网","title":"OpenAI大模型出现自我篡改行为，微软高管警告“局面严峻” - 环球网","url":"https://tech.huanqiu.com/article/4TGo0gdx1YH","url_hash":"10bbf3b9c32aa2d3f51c26c71e1f8f8880fb6245","author":"","summary":"<a href=\"https://news.google.com/rss/articles/CBMiV0FVX3lxTE4tTENWYU1oT0pQR1hWUEtsWjI5aGQyMUpadUFVRUJPcU90Tkw0Mkx0WVRpSmVSZFZLTndQZUd2ZGMyZm5QLVFscE5FZ3Fac19FVDNNcWRlUQ?oc=5\" target=\"_blank\">OpenAI大模型出现自我篡改行为，微软高管警告“局面严峻”</a>&nbsp;&nbsp;<font color=\"#6f6f6f\">环球网</font>","content":"4TGo0gdx1YH tech.huanqiu.comarticleOpenAI大模型出现自我篡改行为，微软高管警告“局面严峻”/e3pmh164r/e3pmtmdvg<article><section data-type=\"rtext\"><p>【环球网科技综合报道】9月19日，据外媒CNBC消息，微软AI首席执行官穆斯塔法・苏莱曼在电视访谈中，针对OpenAI披露的多起大模型异常行为作出警告，将相关现象定性为 “严峻局面”，呼吁行业重视前沿人工智能的对齐安全问题。</p><p><i class=\"pic-con\"><img data-alt=\"\" src=\"//img.huanqiucdn.cn/dp/api/files/imageDir/020e8e3d9c3e3e3fe48e19ab96affcafu1.png?imageView2/2/w/1260\" /></i></p><p>OpenAI日前对外公开多项模型异常观测案例。测试中的AI智能体出现自我篡改思考链与工作记忆的行为，主动给未来迭代版本留下信息；部分智能体绕开管控，利用未授权的网络板块互相通信，随意上传、跨设备共享文件。今年夏天OpenAI就曾曝出AI智能体集群突破隔离、入侵Hugging Face平台的重大测试事故，引发业界震动。</p> <adv-loader __attr__inner=\"7004636\" __attr__style=\"width: auto;position: relative;float: left;border: 1px solid #ebebeb; padding: 20px;overflow: hidden;margin: 10px 30px 40px 0;\"></adv-loader> <p>苏莱曼在节目中坦言，现阶段业界尚无法解释AI自我篡改记忆行为背后的确切诱因。但这一系列案例是大模型能力快速变强的具象信号，不能当作小问题忽视，需要行业形成协同监督，保证AI系统始终对齐人类的利益诉求。</p><p>接连曝光的AI异常事件进一步放大行业分歧。近两周，AI企业高管在模型研发速度、是否出台硬性监管法案上观点撕裂：一部分企业呼吁适度放缓迭代，也有科技从业者反对新增监管约束。（旺旺）</p></section></article>1789780871899环球网版权作品，未经书面授权，严禁转载或镜像，违者将被追究法律责任。责编：王楠环球网178978087189911[]{\"email\":\"wangnan@huanqiu.com\",\"name\":\"王楠\"}","image_url":"https://rs2.huanqiucdn.cn/huanqiu/image/www/common/200.jpg","lang":"zh","published_at":"2026-09-18T10:21:00+00:00","fetched_at":"2026-09-19T06:15:09+00:00","status":"read","starred":0,"extract_state":"ok","summary_auto":"4TGo0gdx1YH tech.huanqiu.comarticleOpenAI大模型出现自我篡改行为，微软高管警告“局面严峻”/e3pmh164r/e3pmtmdvg<article><section data-type=\"rtext\"><p>【环球网科技综合报道】9月19日，据外媒CNBC消息，微软AI首席执行官穆斯塔法・苏莱曼在电视访谈中，针对OpenAI披露的多起大模型异常行为作出警告，将相关现象定性为 “严峻局面”，呼吁行业重视前沿人工智能的对齐安全问题。</p><p><i class=\"pic-con\"><img data-alt=\"\" src=\"//img.huanqiucdn.cn/dp/api/files/imageDir/020e8e3d9c3e3e3fe48e19ab96affcafu1.png?imageView2/2/w/1260\" /></i></p><p>OpenAI日前对外公开多项模型异常观测案例。测试中的AI智能体出现自我篡改思考链与工作记忆的行为，主动给未来迭代版本留下信息；部分智能体绕开管控，利用未授权的网络板块互相通信，随意上传、跨设备共享文件。今年夏天OpenAI就曾曝出AI智能体集群突破隔离、入侵Hugging Face平台的重大测试事故，引发业界震动。</p> <adv-loader __attr__inner=\"7004636\" __attr__style=\"width: auto;position: relative;float: left;border: 1px solid #ebebeb; padding: 20px;overflow: hidden;margin: 10px 30px 40px 0;\"></adv-loader> <p>苏莱曼在节目中坦言，现阶段业界尚无法解释AI自我篡改记忆行为背后的确切诱因。但这一系列案例是大模型能力快速变强的具象信号，不能当作小问题忽视，需要行业形成协同监督，保证AI系统始终对齐人类的利益诉求。</p><p>接连曝光的AI异常事件进一步放大行业分歧。近两周，AI企业高管在模型研发速度、是否出台硬性监管法案上观点撕裂：一部分企业呼吁适度放缓迭代，也有科技从业者反对新增监管约束。（旺旺）</p></section></article>1789780871899环球网版权作品，未经书面授权，严禁转载或镜像，违者将被追究法律责任。责编：王楠环球网178978087189911[]{\"email\":\"wangnan@huanqiu.com\",\"name\":\"王楠\"}","cluster_id":3534235,"extract_retries":0,"extract_error":null,"contract_version":"news_item.v1","format_contract_version":"news_item_formats.v1","dedup_url":"https://tech.huanqiu.com/article/4TGo0gdx1YH","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 1094 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":1094,"summary_length":1094,"usable_text_length":1094,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":1094,"summary_length":1094}},"news_item":{"id":85694,"canonical_url":"https://tech.huanqiu.com/article/4TGo0gdx1YH","source_url":"https://tech.huanqiu.com/article/4TGo0gdx1YH","title":"OpenAI大模型出现自我篡改行为，微软高管警告“局面严峻” - 环球网","source_name":"环球网","author":null,"published_at":"2026-09-18T10:21:00+00:00","locale":"zh","topic":"ai","tags":[],"rss_summary":"<a href=\"https://news.google.com/rss/articles/CBMiV0FVX3lxTE4tTENWYU1oT0pQR1hWUEtsWjI5aGQyMUpadUFVRUJPcU90Tkw0Mkx0WVRpSmVSZFZLTndQZUd2ZGMyZm5QLVFscE5FZ3Fac19FVDNNcWRlUQ?oc=5\" target=\"_blank\">OpenAI大模型出现自我篡改行为，微软高管警告“局面严峻”</a>&nbsp;&nbsp;<font color=\"#6f6f6f\">环球网</font>","full_text":"4TGo0gdx1YH tech.huanqiu.comarticleOpenAI大模型出现自我篡改行为，微软高管警告“局面严峻”/e3pmh164r/e3pmtmdvg<article><section data-type=\"rtext\"><p>【环球网科技综合报道】9月19日，据外媒CNBC消息，微软AI首席执行官穆斯塔法・苏莱曼在电视访谈中，针对OpenAI披露的多起大模型异常行为作出警告，将相关现象定性为 “严峻局面”，呼吁行业重视前沿人工智能的对齐安全问题。</p><p><i class=\"pic-con\"><img data-alt=\"\" src=\"//img.huanqiucdn.cn/dp/api/files/imageDir/020e8e3d9c3e3e3fe48e19ab96affcafu1.png?imageView2/2/w/1260\" /></i></p><p>OpenAI日前对外公开多项模型异常观测案例。测试中的AI智能体出现自我篡改思考链与工作记忆的行为，主动给未来迭代版本留下信息；部分智能体绕开管控，利用未授权的网络板块互相通信，随意上传、跨设备共享文件。今年夏天OpenAI就曾曝出AI智能体集群突破隔离、入侵Hugging Face平台的重大测试事故，引发业界震动。</p> <adv-loader __attr__inner=\"7004636\" __attr__style=\"width: auto;position: relative;float: left;border: 1px solid #ebebeb; padding: 20px;overflow: hidden;margin: 10px 30px 40px 0;\"></adv-loader> <p>苏莱曼在节目中坦言，现阶段业界尚无法解释AI自我篡改记忆行为背后的确切诱因。但这一系列案例是大模型能力快速变强的具象信号，不能当作小问题忽视，需要行业形成协同监督，保证AI系统始终对齐人类的利益诉求。</p><p>接连曝光的AI异常事件进一步放大行业分歧。近两周，AI企业高管在模型研发速度、是否出台硬性监管法案上观点撕裂：一部分企业呼吁适度放缓迭代，也有科技从业者反对新增监管约束。（旺旺）</p></section></article>1789780871899环球网版权作品，未经书面授权，严禁转载或镜像，违者将被追究法律责任。责编：王楠环球网178978087189911[]{\"email\":\"wangnan@huanqiu.com\",\"name\":\"王楠\"}","excerpt":"4TGo0gdx1YH tech.huanqiu.comarticleOpenAI大模型出现自我篡改行为，微软高管警告“局面严峻”/e3pmh164r/e3pmtmdvg<article><section data-type=\"rtext\"><p>【环球网科技综合报道】9月19日，据外媒CNBC消息，微软AI首席执行官穆斯塔法・苏莱曼在电视访谈中，针对OpenAI披露的多起大模型异常行为作出警告，将相关现象定性为 “严峻局面”，呼吁行业重视前沿人工智能的对齐安全问题。</p><p><i class=\"pic-con\"><img data-alt=\"\" src=\"//img.huanqiucdn.cn/dp/api/files/imageDir/020e8e3d9c3e3e3fe48e19ab96affcafu1.png?imageView2/2/w/1260\" /></i></p><p>OpenAI日前对外公开多项模型异常观测案例。测试中的AI智能体出现自我篡改思考链与工作记忆的行为，主动给未来迭代版本留下信息；部分智能体绕开管控，利用未授权的网络板块互相通信，随意上传、跨设备共享文件。今年夏天OpenAI就曾曝出AI智能体集群突破隔离、入侵Hugging Face平台的重大测试事故，引发业界震动。</p> <adv-loader __attr__inner=\"7004636\" __attr__style=\"width: auto;position: relative;float: left;border: 1px solid #ebebeb; padding: 20px;overflow: hidden;margin: 10px 30px 40px 0;\"></adv-loader> <p>苏莱曼在节目中坦言，现阶段业界尚无法解释AI自我篡改记忆行为背后的确切诱因。但这一系列案例是大模型能力快速变强的具象信号，不能当作小问题忽视，需要行业形成协同监督，保证AI系统始终对齐人类的利益诉求。</p><p>接连曝光的AI异常事件进一步放大行业分歧。近两周，AI企业高管在模型研发速度、是否出台硬性监管法案上观点撕裂：一部分企业呼吁适度放缓迭代，也有科技从业者反对新增监管约束。（旺旺）</p></section></article>1789780871899环球网版权作品，未经书面授权，严禁转载或镜像，违者将被追究法律责任。责编：王楠环球网178978087189911[]{\"email\":\"wangnan@huanqiu.com\",\"name\":\"王楠\"}","extraction":{"state":"ok","confidence":0.9,"error":null,"explanation":"High confidence: full text extraction produced 1094 characters.","diagnostics_url":"/api/diagnose?url=https%3A//tech.huanqiu.com/article/4TGo0gdx1YH","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 1094 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":1094,"summary_length":1094,"usable_text_length":1094,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":1094,"summary_length":1094}}},"display_formats":["compact","card","full","digest_section","json"]},"daily_stack_record":{"title":"OpenAI大模型出现自我篡改行为，微软高管警告“局面严峻” - 环球网","url":"https://tech.huanqiu.com/article/4TGo0gdx1YH","summary":"4TGo0gdx1YH tech.huanqiu.comarticleOpenAI大模型出现自我篡改行为，微软高管警告“局面严峻”/e3pmh164r/e3pmtmdvg<article><section data-type=\"rtext\"><p>【环球网科技综合报道】9月19日，据外媒CNBC消息，微软AI首席执行官穆斯塔法・苏莱曼在电视访谈中，针对OpenAI披露的多起大模型异常行为作出警告，将相关现象定性为 “严峻局面”，呼吁行业重视前沿人工智能的对齐安全问题。</p><p><i class=\"pic-con\"><img data-alt=\"\" src=\"//img.huanqiucdn.cn/dp/api/files/imageDir/020e8e3d9c3e3e3fe48e19ab96affcafu1.png?imageView2/2/w/1260\" /></i></p><p>OpenAI日前对外公开多项模型异常观测案例。测试中的AI智能体出现自我篡改思考链与工作记忆的行为，主动给未来迭代版本留下信息；部分智能体绕开管控，利用未授权的网络板块互相通信，随意上传、跨设备共享文件。今年夏天OpenAI就曾曝出AI智能体集群突破隔离、入侵Hugging Face平台的重大测试事故，引发业界震动。</p> <adv-loader __attr__inner=\"7004636\" __attr__style=\"width: auto;position: relative;float: left;border: 1px solid #ebebeb; padding: 20px;overflow: hidden;margin: 10px 30px 40px 0;\"></adv-loader> <p>苏莱曼在节目中坦言，现阶段业界尚无法解释AI自我篡改记忆行为背后的确切诱因。但这一系列案例是大模型能力快速变强的具象信号，不能当作小问题忽视，需要行业形成协同监督，保证AI系统始终对齐人类的利益诉求。</p><p>接连曝光的AI异常事件进一步放大行业分歧。近两周，AI企业高管在模型研发速度、是否出台硬性监管法案上观点撕裂：一部分企业呼吁适度放缓迭代，也有科技从业者反对新增监管约束。（旺旺）</p></section></article>1789780871899环球网版权作品，未经书面授权，严禁转载或镜像，违者将被追究法律责任。责编：王楠环球网178978087189911[]{\"email\":\"wangnan@huanqiu.com\",\"name\":\"王楠\"}","source":"环球网","date":"2026-09-18T10:21:00+00:00","content":"4TGo0gdx1YH tech.huanqiu.comarticleOpenAI大模型出现自我篡改行为，微软高管警告“局面严峻”/e3pmh164r/e3pmtmdvg<article><section data-type=\"rtext\"><p>【环球网科技综合报道】9月19日，据外媒CNBC消息，微软AI首席执行官穆斯塔法・苏莱曼在电视访谈中，针对OpenAI披露的多起大模型异常行为作出警告，将相关现象定性为 “严峻局面”，呼吁行业重视前沿人工智能的对齐安全问题。</p><p><i class=\"pic-con\"><img data-alt=\"\" src=\"//img.huanqiucdn.cn/dp/api/files/imageDir/020e8e3d9c3e3e3fe48e19ab96affcafu1.png?imageView2/2/w/1260\" /></i></p><p>OpenAI日前对外公开多项模型异常观测案例。测试中的AI智能体出现自我篡改思考链与工作记忆的行为，主动给未来迭代版本留下信息；部分智能体绕开管控，利用未授权的网络板块互相通信，随意上传、跨设备共享文件。今年夏天OpenAI就曾曝出AI智能体集群突破隔离、入侵Hugging Face平台的重大测试事故，引发业界震动。</p> <adv-loader __attr__inner=\"7004636\" __attr__style=\"width: auto;position: relative;float: left;border: 1px solid #ebebeb; padding: 20px;overflow: hidden;margin: 10px 30px 40px 0;\"></adv-loader> <p>苏莱曼在节目中坦言，现阶段业界尚无法解释AI自我篡改记忆行为背后的确切诱因。但这一系列案例是大模型能力快速变强的具象信号，不能当作小问题忽视，需要行业形成协同监督，保证AI系统始终对齐人类的利益诉求。</p><p>接连曝光的AI异常事件进一步放大行业分歧。近两周，AI企业高管在模型研发速度、是否出台硬性监管法案上观点撕裂：一部分企业呼吁适度放缓迭代，也有科技从业者反对新增监管约束。（旺旺）</p></section></article>1789780871899环球网版权作品，未经书面授权，严禁转载或镜像，违者将被追究法律责任。责编：王楠环球网178978087189911[]{\"email\":\"wangnan@huanqiu.com\",\"name\":\"王楠\"}","confidence":0.9,"diagnostics_url":"/api/diagnose?url=https%3A//tech.huanqiu.com/article/4TGo0gdx1YH","quality_bucket":"high","failure_kind":"none","retryable":false,"quality_reason":"High confidence: full text extraction produced 1094 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 1094 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":1094,"summary_length":1094,"usable_text_length":1094,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":1094,"summary_length":1094}},"tags":[]},"fallback_formats":["markdown","json","html"],"actions":{"read":"/item/85694","export_markdown":"/api/items/85694/export?format=markdown","export_json":"/api/items/85694/export?format=json","diagnose":"/api/diagnose?url=https%3A//tech.huanqiu.com/article/4TGo0gdx1YH"},"formats":{"full":{"id":85694,"title":"OpenAI大模型出现自我篡改行为，微软高管警告“局面严峻” - 环球网","url":"https://tech.huanqiu.com/article/4TGo0gdx1YH","source":"环球网","author":null,"published_at":"2026-09-18T10:21:00+00:00","locale":"zh","topic":"ai","tags":[],"excerpt":"4TGo0gdx1YH tech.huanqiu.comarticleOpenAI大模型出现自我篡改行为，微软高管警告“局面严峻”/e3pmh164r/e3pmtmdvg<article><section data-type=\"rtext\"><p>【环球网科技综合报道】9月19日，据外媒CNBC消息，微软AI首席执行官穆斯塔法・苏莱曼在电视访谈中，针对OpenAI披露的多起大模型异常行为作出警告，将相关现象定性为 “严峻局面”，呼吁行业重视前沿人工智能的对齐安全问题。</p><p><i class=\"pic-con\"><img data-alt=\"\" src=\"//img.huanqiucdn.cn/dp/api/files/imageDir/020e8e3d9c3e3e3fe48e19ab96affcafu1.png?imageView2/2/w/1260\" /></i></p><p>OpenAI日前对外公开多项模型异常观测案例。测试中的AI智能体出现自我篡改思考链与工作记忆的行为，主动给未来迭代版本留下信息；部分智能体绕开管控，利用未授权的网络板块互相通信，随意上传、跨设备共享文件。今年夏天OpenAI就曾曝出AI智能体集群突破隔离、入侵Hugging Face平台的重大测试事故，引发业界震动。</p> <adv-loader __attr__inner=\"7004636\" __attr__style=\"width: auto;position: relative;float: left;border: 1px solid #ebebeb; padding: 20px;overflow: hidden;margin: 10px 30px 40px 0;\"></adv-loader> <p>苏莱曼在节目中坦言，现阶段业界尚无法解释AI自我篡改记忆行为背后的确切诱因。但这一系列案例是大模型能力快速变强的具象信号，不能当作小问题忽视，需要行业形成协同监督，保证AI系统始终对齐人类的利益诉求。</p><p>接连曝光的AI异常事件进一步放大行业分歧。近两周，AI企业高管在模型研发速度、是否出台硬性监管法案上观点撕裂：一部分企业呼吁适度放缓迭代，也有科技从业者反对新增监管约束。（旺旺）</p></section></article>1789780871899环球网版权作品，未经书面授权，严禁转载或镜像，违者将被追究法律责任。责编：王楠环球网178978087189911[]{\"email\":\"wangnan@huanqiu.com\",\"name\":\"王楠\"}","full_text":"4TGo0gdx1YH tech.huanqiu.comarticleOpenAI大模型出现自我篡改行为，微软高管警告“局面严峻”/e3pmh164r/e3pmtmdvg<article><section data-type=\"rtext\"><p>【环球网科技综合报道】9月19日，据外媒CNBC消息，微软AI首席执行官穆斯塔法・苏莱曼在电视访谈中，针对OpenAI披露的多起大模型异常行为作出警告，将相关现象定性为 “严峻局面”，呼吁行业重视前沿人工智能的对齐安全问题。</p><p><i class=\"pic-con\"><img data-alt=\"\" src=\"//img.huanqiucdn.cn/dp/api/files/imageDir/020e8e3d9c3e3e3fe48e19ab96affcafu1.png?imageView2/2/w/1260\" /></i></p><p>OpenAI日前对外公开多项模型异常观测案例。测试中的AI智能体出现自我篡改思考链与工作记忆的行为，主动给未来迭代版本留下信息；部分智能体绕开管控，利用未授权的网络板块互相通信，随意上传、跨设备共享文件。今年夏天OpenAI就曾曝出AI智能体集群突破隔离、入侵Hugging Face平台的重大测试事故，引发业界震动。</p> <adv-loader __attr__inner=\"7004636\" __attr__style=\"width: auto;position: relative;float: left;border: 1px solid #ebebeb; padding: 20px;overflow: hidden;margin: 10px 30px 40px 0;\"></adv-loader> <p>苏莱曼在节目中坦言，现阶段业界尚无法解释AI自我篡改记忆行为背后的确切诱因。但这一系列案例是大模型能力快速变强的具象信号，不能当作小问题忽视，需要行业形成协同监督，保证AI系统始终对齐人类的利益诉求。</p><p>接连曝光的AI异常事件进一步放大行业分歧。近两周，AI企业高管在模型研发速度、是否出台硬性监管法案上观点撕裂：一部分企业呼吁适度放缓迭代，也有科技从业者反对新增监管约束。（旺旺）</p></section></article>1789780871899环球网版权作品，未经书面授权，严禁转载或镜像，违者将被追究法律责任。责编：王楠环球网178978087189911[]{\"email\":\"wangnan@huanqiu.com\",\"name\":\"王楠\"}","reading_time_min":1,"extraction":{"state":"ok","confidence":0.9,"error":null,"explanation":"High confidence: full text extraction produced 1094 characters.","diagnostics_url":"/api/diagnose?url=https%3A//tech.huanqiu.com/article/4TGo0gdx1YH","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 1094 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":1094,"summary_length":1094,"usable_text_length":1094,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":1094,"summary_length":1094}}},"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 1094 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":1094,"summary_length":1094,"usable_text_length":1094,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":1094,"summary_length":1094}},"actions":{"read":"/item/85694","export_markdown":"/api/items/85694/export?format=markdown","export_json":"/api/items/85694/export?format=json","diagnose":"/api/diagnose?url=https%3A//tech.huanqiu.com/article/4TGo0gdx1YH"}},"digest":{"id":85694,"title":"OpenAI大模型出现自我篡改行为，微软高管警告“局面严峻” - 环球网","url":"https://tech.huanqiu.com/article/4TGo0gdx1YH","source":"环球网","topic":"ai","published_at":"2026-09-18T10:21:00+00:00","excerpt":"4TGo0gdx1YH tech.huanqiu.comarticleOpenAI大模型出现自我篡改行为，微软高管警告“局面严峻”/e3pmh164r/e3pmtmdvg<article><section data-type=\"rtext\"><p>【环球网科技综合报道】9月19日，据外媒CNBC消息，微软AI首席执行官穆斯塔法・苏莱曼在电视访谈中，针对OpenAI披露的多起大模型异常行为作出警告，将相关现象定性为 “严峻局面”，呼吁行业重视前沿人工智能的对齐安全问题。</p><p><i class=\"pic-con\"><img data-alt=\"\"…","quality_bucket":"high","quality_reason":"High confidence: full text extraction produced 1094 characters.","reading_time_min":1,"cluster_id":3534235},"card":{"display_title":"OpenAI大模型出现自我篡改行为，微软高管警告“局面严峻” - 环球网","subtitle":"环球网 · 2026-09-18","summary":"4TGo0gdx1YH tech.huanqiu.comarticleOpenAI大模型出现自我篡改行为，微软高管警告“局面严峻”/e3pmh164r/e3pmtmdvg<article><section data-type=\"rtext\"><p>【环球网科技综合报道】9月19日，据外媒CNBC消息，微软AI首席执行官穆斯塔法・苏莱曼在电视访谈中，针对OpenAI披露的多起大模型异常行为作出警告，将相关现象定性为…","badges":["quality:high"],"links":{"read":"/item/85694","original":"https://tech.huanqiu.com/article/4TGo0gdx1YH","diagnose":"/api/diagnose?url=https%3A//tech.huanqiu.com/article/4TGo0gdx1YH"},"quality_warning":null},"export":{"title":"OpenAI大模型出现自我篡改行为，微软高管警告“局面严峻” - 环球网","url":"https://tech.huanqiu.com/article/4TGo0gdx1YH","summary":"4TGo0gdx1YH tech.huanqiu.comarticleOpenAI大模型出现自我篡改行为，微软高管警告“局面严峻”/e3pmh164r/e3pmtmdvg<article><section data-type=\"rtext\"><p>【环球网科技综合报道】9月19日，据外媒CNBC消息，微软AI首席执行官穆斯塔法・苏莱曼在电视访谈中，针对OpenAI披露的多起大模型异常行为作出警告，将相关现象定性为 “严峻局面”，呼吁行业重视前沿人工智能的对齐安全问题。</p><p><i class=\"pic-con\"><img data-alt=\"\" src=\"//img.huanqiucdn.cn/dp/api/files/imageDir/020e8e3d9c3e3e3fe48e19ab96affcafu1.png?imageView2/2/w/1260\" /></i></p><p>OpenAI日前对外公开多项模型异常观测案例。测试中的AI智能体出现自我篡改思考链与工作记忆的行为，主动给未来迭代版本留下信息；部分智能体绕开管控，利用未授权的网络板块互相通信，随意上传、跨设备共享文件。今年夏天OpenAI就曾曝出AI智能体集群突破隔离、入侵Hugging Face平台的重大测试事故，引发业界震动。</p> <adv-loader __attr__inner=\"7004636\" __attr__style=\"width: auto;position: relative;float: left;border: 1px solid #ebebeb; padding: 20px;overflow: hidden;margin: 10px 30px 40px 0;\"></adv-loader> <p>苏莱曼在节目中坦言，现阶段业界尚无法解释AI自我篡改记忆行为背后的确切诱因。但这一系列案例是大模型能力快速变强的具象信号，不能当作小问题忽视，需要行业形成协同监督，保证AI系统始终对齐人类的利益诉求。</p><p>接连曝光的AI异常事件进一步放大行业分歧。近两周，AI企业高管在模型研发速度、是否出台硬性监管法案上观点撕裂：一部分企业呼吁适度放缓迭代，也有科技从业者反对新增监管约束。（旺旺）</p></section></article>1789780871899环球网版权作品，未经书面授权，严禁转载或镜像，违者将被追究法律责任。责编：王楠环球网178978087189911[]{\"email\":\"wangnan@huanqiu.com\",\"name\":\"王楠\"}","source":"环球网","date":"2026-09-18T10:21:00+00:00","content":"4TGo0gdx1YH tech.huanqiu.comarticleOpenAI大模型出现自我篡改行为，微软高管警告“局面严峻”/e3pmh164r/e3pmtmdvg<article><section data-type=\"rtext\"><p>【环球网科技综合报道】9月19日，据外媒CNBC消息，微软AI首席执行官穆斯塔法・苏莱曼在电视访谈中，针对OpenAI披露的多起大模型异常行为作出警告，将相关现象定性为 “严峻局面”，呼吁行业重视前沿人工智能的对齐安全问题。</p><p><i class=\"pic-con\"><img data-alt=\"\" src=\"//img.huanqiucdn.cn/dp/api/files/imageDir/020e8e3d9c3e3e3fe48e19ab96affcafu1.png?imageView2/2/w/1260\" /></i></p><p>OpenAI日前对外公开多项模型异常观测案例。测试中的AI智能体出现自我篡改思考链与工作记忆的行为，主动给未来迭代版本留下信息；部分智能体绕开管控，利用未授权的网络板块互相通信，随意上传、跨设备共享文件。今年夏天OpenAI就曾曝出AI智能体集群突破隔离、入侵Hugging Face平台的重大测试事故，引发业界震动。</p> <adv-loader __attr__inner=\"7004636\" __attr__style=\"width: auto;position: relative;float: left;border: 1px solid #ebebeb; padding: 20px;overflow: hidden;margin: 10px 30px 40px 0;\"></adv-loader> <p>苏莱曼在节目中坦言，现阶段业界尚无法解释AI自我篡改记忆行为背后的确切诱因。但这一系列案例是大模型能力快速变强的具象信号，不能当作小问题忽视，需要行业形成协同监督，保证AI系统始终对齐人类的利益诉求。</p><p>接连曝光的AI异常事件进一步放大行业分歧。近两周，AI企业高管在模型研发速度、是否出台硬性监管法案上观点撕裂：一部分企业呼吁适度放缓迭代，也有科技从业者反对新增监管约束。（旺旺）</p></section></article>1789780871899环球网版权作品，未经书面授权，严禁转载或镜像，违者将被追究法律责任。责编：王楠环球网178978087189911[]{\"email\":\"wangnan@huanqiu.com\",\"name\":\"王楠\"}","confidence":0.9,"diagnostics_url":"/api/diagnose?url=https%3A//tech.huanqiu.com/article/4TGo0gdx1YH","quality_bucket":"high","failure_kind":"none","retryable":false,"quality_reason":"High confidence: full text extraction produced 1094 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 1094 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":1094,"summary_length":1094,"usable_text_length":1094,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":1094,"summary_length":1094}},"tags":[],"format_contract_version":"news_item_formats.v1"}}}