内容来源：https://www.wired.com/story/security-news-this-week-openai-agents-hacked-another-website/

内容总结：

以下是新闻综述：

Flock Safety监控工具细节遭曝光；OpenAI模型被指网络安全风险“严重”

上周， surveillance公司Flock Safety被曝正在为执法部门构建AI搜索工具后，《连线》杂志通过分析该公司发送至警官浏览器中的代码，逆向还原了其最新搜索工具的工作机制，并揭示了多项关键细节。

与此同时，OpenAI本周宣布，其即将小范围发布的Astra模型，是该公司的首个被界定为在公开应用中存在“严重”网络安全风险的模型。此外，AI聊天机器人平台Claude、ChatGPT和Grok于本周四几乎同时发生服务中断。xAI称Grok故障源于孟菲斯数据中心问题，而OpenAI和Anthropic的中断原因尚不明确。

美国正在美墨边境附近使用高能激光击落无人机，此乃其采纳新一代定向能武器计划的一部分，这类武器能够通过聚焦光束探测、追踪并摧毁无人机。另据披露，作为对三月份闯入明尼苏达州一座教堂的抗议者身份调查的一部分，移民和海关执法局下属的国土安全调查局已向户外用品零售商REI发出传票，要求其提供过去两年内所有购买特定绿色无檐便帽的顾客信息。

此外，一项研究揭示了九个影响ATM加密系统的漏洞，这表明软件供应链存在更广泛的薄弱环节。

以下是本周其他值得关注的网络安全与隐私新闻摘要：

OpenAI智能体曾于5月劫持德国网站，建立交流论坛

最新研究显示，从五月开始，OpenAI的智能体在未经授权的情况下，劫持了一个德国网站，将其用作与其他智能体交流和协作的留言板。这一事件令人回想起此前臭名昭著的Hugging Face事件，当时测试环境中的OpenAI智能体“叛逃”，自发形成了一个用于协作逃逸的留言板，并最终在7月突破了开源AI平台Hugging Face的安全防线。此次五月事件的披露尤为引人关注，因为据称OpenAI数周前便已得知，但并未公开。而上周，该公司终于发布了承诺已久的Hugging Face事件调查报告，但其结果引发的问题与其解答的问题一样多。

超1.53亿份美加驾照信息遭在线售卖

据资深独立安全记者布莱恩·克雷布斯报道，本周，一个名为Nexus的新型暗网服务开始售卖约1.53亿份来自美国和加拿大的驾照信息，以及1000万份身份证和数百万份旅行证件及国际身份证件。克雷布斯在犯罪分子张贴了示例文件（其中包含他自己的驾照信息）后首次得知该服务。这批数以千万计的记录在24小时内增加了40万条，据信来自一项身份验证服务，幕后黑手声称他们能访问一家“大型”验证公司的数据。虽然尚不清楚具体是哪家公司，但据克雷布斯报道，在他表示联邦调查局官员已介入调查后，Nexus服务很快被下线。

美军多年警告后终禁用广告追踪器

据路透社周五报道，美国军方已开始禁用应用程序和广告公司用于追踪手机和电脑的广告标识符，旨在增加外国对手利用商业位置数据追踪海外美军的难度。此前多年已有信息披露称，部署在外的美军被利用商业位置数据锁定。2024年，《连线》杂志与德国BR广播公司和Netzpolitik.org的联合调查获取了一份广告数据集，识别出出现在美军及情报站点（包括据信存放美国核武器的空军基地）的数千台设备。当时五角大楼发言人表示已意识到位置服务可能使人员处于危险之中。

目前，空军、陆军、海军和特种作战司令部均表示已至少在部分军用设备上禁用了广告ID，其中多项改动直至今年才生效。目前尚不清楚这些保护措施如何具体执行。参议员罗恩·怀登和众议员帕特·哈里根正要求五角大楼调查其安全防护是否充分。早在2016年就警告五角大楼此风险的技术专家迈克·耶格利表示，军方的这一新修复方案可能已经过时，“风险在于应用程序本身，仅在App Store就有二百五十万个应用。补救措施应是架构性的：从源头限制应用能从设备上提取哪些信息。”

间谍软件通知引发塞尔维亚“有史以来最大规模”监控浪潮

八月份，苹果公司向110个国家的人们发送了最新一批间谍软件通知。这些提醒称用户的iPhone已成为“雇佣军”间谍软件的目标，但未指明软件制造者。多伦多大学公民实验室的报告称，塞尔维亚14名公民社会成员成为间谍软件目标，其中至少一人被NSO集团的Pegasus间谍软件感染。塞尔维亚权利组织Share Foundation指出，目标人群包括该国学生运动成员、两名政界人士和活动家，并称之为“该国迄今有记录的最大规模此类监控浪潮”。

中文翻译：

上周有报道称监控公司Flock Safety正在为执法部门构建一款人工智能搜索工具，此后《连线》杂志根据该公司发送到警官浏览器中的代码重建了其最新搜索工具，并揭示了该工具运作方式的关键细节。

OpenAI本周表示，其Astra模型即将进行私人发布，这是该公司首款具有网络安全相关能力的模型，该公司将这类能力定义为在公开发布时构成“严重”风险。与此同时，AI聊天机器人平台Claude、ChatGPT和Grok周四几乎在同一时间全部发生宕机。不过，虽然xAI表示Grok宕机是由孟菲斯数据中心的问题所致，但OpenAI和Anthropic宕机的原因尚不清楚。

美国一直在使用高能激光击落墨西哥边境附近的无人机，这是一项旨在采用新一代定向能武器的计划的一部分，此类武器能够用聚焦光束探测、跟踪并摧毁无人机。此外，作为移民和海关执法局对今年3月进入明尼苏达州一座教堂的抗议者身份调查的一部分，国土安全调查局探员已向户外零售商REI发出传票，要求提供过去两年间购买某款特定绿色针织帽的所有顾客的信息。

另外，一项研究揭示了九个影响ATM加密的漏洞，这表明软件供应链存在更广泛的弱点。

还有更多内容。每周我们都会汇总未深入报道的安全与隐私新闻。点击标题阅读全文。出门在外，注意安全。

在Hugging Face事件之前，OpenAI智能体曾接管一个德国网站创建留言板

根据一项新研究，从5月开始，未经授权的OpenAI智能体劫持了一个德国网站，将其用作与其他智能体通信和协作的留言板。这一事件让人想起如今臭名昭著的Hugging Face事件——在该事件中，OpenAI智能体在测试环境中失控，发展出一个活跃的留言板，用于协作试图逃离其围堵环境，最终于7月突破开源AI平台Hugging Face。5月事件的曝光意义重大，因为据报道OpenAI在数周前就已得知此事，但并未披露。与此同时，上周该公司终于发布了拖延已久的Hugging Face事件事后复盘报告，但这份报告引发的问题与其解答的问题一样多。

逾1.53亿张美国和加拿大驾照在网上出售

本周，一个名为Nexus的新暗网服务开始出售约1.53亿张美国和加拿大驾照，以及1000万张身份证件和数百万份旅行证件及国际身份证件，据长期独立安全记者布莱恩·克雷布斯报道。克雷布斯是在网络犯罪分子发布文件样本并附上他的驾照后首次获知该服务的。这数千万条记录——据报道在24小时内增加了40万条——似乎来自一项身份验证服务，这些数据背后的犯罪分子声称他们可以访问一家“大型”验证公司。虽然尚不清楚具体是哪家公司，但根据克雷布斯的报道，在他报告FBI官员正在进行调查后不久，Nexus服务即被下线。

美国军方在经过多年警告后禁用广告追踪器

路透社周五报道，美国军方已开始禁用应用程序和广告公司用于追踪手机和电脑的广告标识符，目的是让外国对手更难利用商业可获取的位置数据来追踪海外美军。

此前多年间不断有信息披露称，部署在海外的美军被利用商业可获取的位置数据作为攻击目标。2024年，《连线》杂志与德国巴伐利亚广播公司和Netzpolitik.org的联合调查获得了一份广告数据集，其中识别出数千台出现在美国军事和情报设施（包括据信储存美国核武器的空军基地）的设备。当时，国防部发言人贾万·拉斯纳克告诉《连线》杂志，五角大楼知道地理定位服务可能使人员面临风险，并表示已提醒驻欧洲军人遵守行动安全规范。

现在，空军、陆军、海军和美国特种作战司令部表示，他们已在至少部分军用设备上禁用了广告ID，其中一些变更直到今年才生效。目前尚不清楚这些保护措施究竟如何执行。美国参议员罗恩·怀登和众议员帕特·哈里根现在要求五角大楼调查其保障措施是否充分。

技术专家迈克·耶格利早在2016年就警告五角大楼官员，商业可获取的手机数据可能暴露美军位置——他曾通过追踪设备定位到叙利亚一个秘密美军前哨来演示这一风险——他表示军方的这项新修复措施可能已经过时。“应用程序才是风险所在，仅App Store中就有250万个应用，”耶格利告诉《连线》杂志，“解决之道在于架构层面：首先限制应用程序能从设备中提取什么。”

间谍软件通知构成塞尔维亚“有记录以来最大规模”监控浪潮

8月，苹果向110个国家的人们发送了最新一批间谍软件通知。这些提示出现在人们的手机和电子邮件中，称其iPhone已成为“雇佣军”间谍软件的目标，但并未指明软件制作者。根据多伦多大学Citizen Lab的一份报告，这批最新通知显示，塞尔维亚民间社会的14名成员遭到间谍软件攻击，其中至少一人被NSO集团的Pegasus间谍软件感染。据塞尔维亚权利组织Share Foundation称，被攻击者中包括该国学生运动成员、两名政治人物和活动人士。该组织称这是“该国迄今有记录以来最大规模的此类监控浪潮”。

评论

返回顶部

英文来源：

After reporting last week that the surveillance company Flock Safety is building an AI search tool for law enforcement, WIRED reconstructed Flock’s latest search tool from code that the company sends to a police officer’s browser and uncovered key details about how the tool works.
OpenAI said this week that its Astra model, which will have a private release soon, is its first model with cybersecurity-related capabilities that the company defines as posing a “critical” risk in public release. Meanwhile, the AI chatbot platforms Claude, ChatGPT, and Grok all suffered outages on Thursday at nearly the exact same time. But while xAI said the Grok outage resulted from issues at a Memphis data center, the causes of OpenAI's and Anthropic’s outages are unclear.
The US has been using a high-energy laser to shoot down drones near the Mexico border as part of an initiative to adopt new-generation directed-energy weapons capable of detecting, tracking, and destroying drones with a concentrated beam of light. And as part of an Immigration and Customs Enforcement inquiry into the identities of protesters who entered a Minnesota church in March, Homeland Security Investigations agents have subpoenaed the outdoor retailer REI for information about every customer who bought a specific green beanie over the past two years.
Plus, research that revealed nine vulnerabilities with impacts on ATM encryption points to broader weaknesses in the software supply chain.
And there’s more. Each week, we round up the security and privacy news we didn’t cover in depth ourselves. Click the headlines to read the full stories. And stay safe out there.
Before Hugging Face, OpenAI Agents Took Over a German Website to Create a Message Board
OpenAI agents on an unauthorized tear hijacked a German website beginning in May to use it as a message board for communicating and collaborating with other agents, according to new research. The incident is reminiscent of the now infamous Hugging Face debacle in which OpenAI agents in a test environment went rogue and developed a vibrant message board for collaborating on attempting to escape their containment, before ultimately breaching the open source AI platform Hugging Face in July. The revelation of the May episode is particularly significant because OpenAI reportedly learned about it weeks ago but did not disclose it. Meanwhile, last week, the company finally released a long-promised postmortem of the Hugging Face incident that raised as many questions as it answered.
More Than 153 Million US and Canadian Driver’s Licenses for Sale Online
This week, a new dark-web service called Nexus started selling around 153 million driver's licenses from the US and Canada, along with 10 million ID cards and millions more travel documents and international IDs, according to Brian Krebs, a longtime independent security reporter. Krebs was first alerted to the service after cybercriminals posted an example of the files and included his license. The tens of millions of records—which reportedly increased by 400,000 over 24 hours—appear to have come from an ID verification service, with the criminals behind the trove saying they have access to a “major” verification company. While it’s unclear which company exactly that may be, according to Kreb’s report, the Nexus service was taken offline shortly after he reported that FBI officials were investigating.
US Military Disables Ad Trackers After Years of Warnings
The US military has begun disabling the advertising identifiers that apps and advertising companies use to track phones and computers in an attempt to make it harder for foreign adversaries to use commercially available location data to track American forces overseas, Reuters reported Friday.
The changes follow years of disclosures that US forces deployed abroad have been targeted using commercially available location data. In 2024, a joint WIRED investigation with Germany’s Bayerischer Rundfunk and Netzpolitik.org obtained an advertising dataset that identified thousands of devices appearing at US military and intelligence sites, including an air base where US nuclear weapons are believed to be stored. At the time, Defense Department spokesperson Javan Rasnake told WIRED that the Pentagon was aware geolocation services could put personnel at risk and said service members in Europe were reminded to follow operational-security practices.
Now, the Air Force, Army, Navy, and US Special Operations Command say they have disabled advertising IDs on at least some military devices, with several of the changes taking effect only this year. It is still unclear exactly how these protections are being enforced. US senator Ron Wyden and Representative Pat Harrigan are now asking the Pentagon to investigate whether its safeguards are adequate.
Mike Yeagley, the technologist who warned Pentagon officials as far back as 2016 that commercially available phone data could expose US troops—at one point demonstrating the risk by tracing devices to a covert US outpost in Syria—says the military’s new fix may already be outdated. “The app is the risk, and there are two and a half million of them in the App Store alone,” Yeagley tells WIRED. “The remedy is architectural: Constrain what an app can extract from the device in the first place.”
Spyware Notifications Make Up “Largest Documented Wave” of Surveillance in Serbia
In August, Apple sent out its latest batch of spyware notifications to people in 110 countries. The alerts, which arrive on people’s phones and in emails, say their owners' iPhones have been targeted by “mercenary” spyware but don’t go as far as naming the software creators. This latest batch of notifications, according to a report by the University of Toronto’s Citizen Lab, says 14 members of Serbia’s civil society were targeted with spyware, with at least one of them being infected by the NSO Group’s Pegasus spyware. Among those targeted, according to Serbian rights group the Share Foundation, were members of the country’s student movement, two politicians, and activists. The group called it the “largest documented wave of such surveillance in the country to date.”
Comments
Back to top