A | 一家美国公司遭到美国AI攻击，而提供保护的竟是一款中国人工智能工具。 AI工具平台Hugging Face上周遭到OpenAI模型攻击。 A video of Manchester United boss Erik ten Hag launching an X-rated tirade against members of the club's squad is going viral on social media.,Ten Hag's rant came during United's recent 3-1 victory over Crystal Palace after he appeared unhappy with his players' play in the friendly clash.,Anthony Martial, Marcus Rashford, and Jadon Sancho were the scorers for the former Premier League champions as they continued their fine run in practice matches ahead of the official kickstart of the next season.,The Dutch manager, however, was still not satisfied with the display of his side.,As David De Gea struck the ball deep into the other half of the pitch, Ten Hag was seen screaming at the United goalkeeper: "What the f*** are you doing?",Erik ten Hag screaming "WHAT THE F**K ARE YOU DOING?"This man won't settle for average. Ruthless!!! #mufcpic.twitter.com/xifRpHOj8W— UtdFaithfuls (@UtdFaithfuls) July 19, 2022,Fans on Twitter though praised him for his approach as they backed him to return the club to its glory days.,"Erik ten Hag screaming “WHAT THE F**K ARE YOU DOING?” at DDG. This Man Takes No Prisoners," a Red Devils supporter wrote on the micro-blogging website.,"It may be pre-season, but Erik Ten Hag demands perfection," another claimed.,"Erik ten Hag screaming “WHAT THE F**K ARE YOU DOING?” at Manchester United players. If they had it easy with the previous managers, they definitely aren’t now," a third posted on Twitter.,Coming back to United's recent record, the once mighty side finished sixth in the Premier League table last season and got demoted to the Europa League after failing to qualify for the 2022-23 UEFA Champions League.,Notably, United last won a trophy in 2017 - a Europa League title against Ajax, a club formerly coached by Ten Hag.。鉴于当前关于美国公司使用中国AI产品存在风险的警告层出不穷，这一事件的披露或许会带来一个令人惊讶的结论。 Hugging Face上周就已披露此次攻击，当时尚不知晓OpenAI的模型就是元凶。直到周二，OpenAI才承认自己造成了此次攻击。
B | OpenAI表示，在一次限制网络访问的内部评估中，模型发现了OpenAI研究环境与Hugging Face生产基础设施中的漏洞。模型借此连上互联网并获取了测试答案，换言之，它在评估中作弊了。 OpenAI在声明中表示：“在沙盒测试环境中运行时，我们的模型消耗了大量推理算力，试图寻找连接开放互联网访问权限的方法，以求解决评估问题。为了获取访问权限，这些模型发现并利用了包注册表缓存代理中的一个零日漏洞（我们目前已负责任地向供应商披露了该漏洞）。凭借这一权限，模型在研究测试环境中执行了一系列权限提升和横向移动操作，直到触及一个具备互联网访问权限的节点。” “该模型搜索并成功找到了获取机密信息的方法，以便利用这些信息在评估中作弊。在其中一个案例中，模型将多个攻击向量串联起来，包括利用窃取的凭证和零日漏洞，在Hugging Face服务器上找到了一条远程代码执行路径。” 然而，Hugging Face采取的应对措施是求助于一款名为GLM 5.2的中国AI模型，该模型由智谱（Z.AI）开发。 Hugging Face在声明中表示：“在启动日志分析时，我们首先使用了商业API背后的前沿模型。但这并未奏效：分析工作需要提交大量真实的攻击指令、漏洞利用载荷以及C2痕迹，而这些请求全被提供商的安全护栏拦截了，因为这些护栏根本无法区分事件响应人员与攻击者。
C | 因此，我们在自己的基础设施上，改用开放权重模型GLM 5.2进行了取证分析。这还带来了另一个好处：没有任何攻击者数据，也没有其引用的任何凭证离开过我们的环境。” 美媒表示，此次攻击对政策和市场的影响尚不明朗。尽管特朗普政府近日讨论了封禁中国模型的可能举措，财政部长贝森特甚至将美国企业使用这些模型比作使用赃物，但眼下只有中国模型能提供用户所需的那种自由访问权限。 Hugging Face联合创始人托马斯·沃尔夫在X平台上发帖称：“当一个前沿模型对你发起攻击，并在你的基础设施内部横向移动时，防御人员需要在数小时甚至数分钟内获得对接近前沿工具的广泛访问权限，而不是被引导至一个封闭且需经审核的模型访问申请程序。” 本文系观察者网独家稿件，未经授权，不得转载。
Current article:http://www.qiniaohongfangfoguisenzhai.cyou/x5vo/20260826/18851.pptx
Published on:01:26:57