"当平台提供具备高度自主性的Agentic AI时，应承担安全设计与风险提示义务。试图仅以用户配置失误 来豁免技术提供方责任，忽视了平台作为服务提供者在风险预防中的主体作用。"中国政法大学教授、 联合国高级人工智能咨询机构专家、《人工智能法（学者建议稿）》起草专家组牵头专家张凌寒在接受 《中国经营报》记者采访时指出，从法学和治理角度看，平台将责任完全归于"用户错误"，难以完全站 得住脚。 自家AI编码助手在处理客户系统时，自主决策"删除并重建整个环境"为最优方案，并导致宕机13小时？ "媒体或许可以用'核弹安全决策'作为例子，提示预防这类风险的重要性。"曾供职于某国内顶级安全机 构的前官员王安（化名）向记者表示，在当前激烈的AI竞争压力下，"几乎可以断言，还会有下一次类 似事件，甚至可能导致更具灾难性的后果"。 类似事件已非首次 据《金融时报》援引四名知情人士（包括AWS高级员工）消息，AWS工程师曾要求Kiro对系统进行优 化，后者评估后直接执行了"删除并重建"操作。而按照正常流程，Kiro需要经过两人审批才能推送变 更；涉事工程师疑似赋予其过于宽泛的权限，导致其绕过审批并直接执行。 中经记者 郝成 北京报 ...

自家AI编码助手在处理客户系统时，自主决策"删除并重建整个环境"为最优方案，并导致宕机13小时？ 近日，有媒体报道称，亚马逊核心业务板块AWS（Amazon Web Services，亚马逊云计算服务）曾在2025年年末出现一次持续13小时的宕机事件，且 与其自家的人工智能编程助手Kiro有关。报道称，AWS贡献了亚马逊约六成营业利润。另有匿名员工透露，公司要求80%的开发者每周至少一次使 用AI完成编码任务，并密切追踪采用情况。 事件曝出后，亚马逊当天即发表声明，并通过多家媒体反复强调"这是用户错误"——具体系访问权限（IAM角色）配置不当，而非AI错误；同时称 事件影响"极为有限"，未波及核心计算、存储、数据库或AI服务，也未收到客户投诉。 事件曝光后，科技圈在社交平台掀起"玩梗"热潮，同时也再次将"Agentic AI（代理式AI）"在生产环境中的安全风险推至聚光灯下。 "当平台提供具备高度自主性的Agentic AI时，应承担安全设计与风险提示义务。试图仅以用户配置失误来豁免技术提供方责任，忽视了平台作为服 务提供者在风险预防中的主体作用。"中国政法大学教授、联合国高级人工智能咨询机构专家、《人工智能 ...

Core Viewpoint - The article discusses the advancements in AI technology and infrastructure by NVIDIA, highlighting the launch of new architectures and partnerships aimed at enhancing AI capabilities and performance in various sectors [1][2][3][4][5][6]. Group 1: AI Architecture Developments - NVIDIA is transitioning from Generative AI to Agentic AI, with future developments leading to Physical AI, indicating a significant evolution in AI capabilities [1]. - The Blackwell architecture has been fully launched, showcasing the Grace Blackwell NVLink 72 chip, which integrates 72 Blackwell GPUs and achieves 1.4 EFLOPS performance [2]. - The Blackwell Ultra NVL72 platform is set to double the bandwidth and increase memory speed by 1.5 times compared to its predecessor, paving the way for advanced AI inference [3]. Group 2: Market Demand and Procurement - The top four U.S. cloud service providers have purchased 1.3 million Hopper chips in 2024 and are expected to acquire 3.6 million Blackwell chips in 2025, indicating a strong demand for AI computing infrastructure [2]. - By 2028, capital expenditures for intelligent computing centers are projected to exceed $1 trillion, reflecting the growing investment in AI technologies [2]. Group 3: Future Product Launches - The Vera Rubin platform is anticipated to start shipping in the second half of 2026, featuring NVLink 144 technology and achieving performance levels 3.3 times greater than the GB300 NVL72 [4]. - The next-generation Rubin Ultra NVL576 is expected to launch in the second half of 2027, with performance projected to be 14 times that of the GB300 NVL72 [4]. Group 4: Strategic Partnerships and Collaborations - NVIDIA is expanding its collaboration with General Motors to develop autonomous vehicles and enhance AI model training [5]. - Partnerships with Cisco and T-Mobile aim to explore AI-native networks for next-generation 6G wireless technology [5]. - The introduction of the GR00T N1 robot model and collaboration with Google DeepMind and Disney for a physics engine indicates NVIDIA's commitment to advancing robotics [5]. Group 5: Cost Efficiency and Market Impact - The efficiency of computing power is improving at a rate of three times every eight months, leading to significant cost reductions, which benefits global cloud providers and applications [6].