研究人员利用OpenAI和谷歌的模型，在全球绝大多数网站使用的软件中发现了严重漏洞。亚马逊警告称，研究人员借助人工智能工具，在支撑互联网大量网站运行的一种编程语言中发现了安全漏洞，目前这些漏洞正遭到“大规模利用”。各国政府和企业正争相采取措施，确保自身系统安全。
这家全球最大的云计算服务提供商周三表示，该公司发现，黑客早在相关漏洞向公众披露的四天前，就已试图利用这些漏洞发动攻击。
亚马逊补充称，这些漏洞“目前正在现实网络环境中遭到广泛利用”。
当天早些时候，美国国家安全局（NSA）和英国有关部门警告称，这些漏洞使全球各地的网站面临极高的黑客攻击风险。据报道，部分欧洲国家的政府已在修补这两处安全漏洞期间，将部分网站暂时下线。
这些漏洞在十多年间一直未被发现，直到研究人员借助OpenAI和谷歌的人工智能模型才将其识别出来。漏洞潜在影响范围之广，凸显了人工智能颠覆网络安全格局的能力。
今年早些时候，Anthropic和OpenAI相继推出具备强大网络安全能力的新模型，引发政界人士和金融监管机构的警告，并促使美国政府暂时限制这些模型的推出。
美国国家安全局周三警告所有使用名为PHP的编程语言的机构，要求其立即更新系统，否则应将相关系统下线。
负责维护PHP的开发人员已于当天早些时候发布软件更新，以修复这些漏洞。
美国国家安全局表示：“这些漏洞往往使攻击者能够直接访问服务器的文件系统或数据库。如果无法立即完成更新，建议相关机构暂时将系统下线，直至漏洞得到彻底修复。”
美国两家网络安全机构Calif和Wiz的研究人员发现了这些漏洞。谷歌母公司Alphabet旗下的Wiz估计，存在漏洞的PHP编程语言支撑着全球约70%的网站运行。
OpenAI周三表示：“鉴于这些漏洞的严重程度以及PHP的广泛应用，我们认为这是近年来影响最为重大的网络安全问题之一。据我们所知，前沿人工智能模型先于攻击者发现了这些漏洞。”
OpenAI补充称：“这凸显了人工智能发现可能影响整个互联网的安全漏洞的能力，但同样的能力也可能帮助攻击者迅速开发攻击工具和漏洞利用程序。”
OpenAI与PHP、Calif和Wiz展开合作，共同验证了这些漏洞，并开发了相应的修复补丁。
白宫表示，联邦机构正在评估这些漏洞可能对政府系统和关键基础设施造成的影响。
PHP之所以在互联网上得到广泛应用，主要是因为全球使用最广泛的网站建设平台WordPress主要采用PHP编写，并依赖这一语言运行。
英国国家网络安全中心（NCSC）表示，这些漏洞可能使黑客无需密码或其他登录凭证，就能控制网站服务器。
该机构表示：“英国国家网络安全中心正敦促英国各机构立即采取行动。”
本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
Using the models of Open AI and Google, researchers found serious gaps in the software used on the vast majority of websites worldwide. The Amazon warns that researchers, with the help of artificial intelligence tools, have found security gaps in a programming language that supports the operation of a large number of Internet sites, which are now being “massively exploited”. Governments and businesses are competing to take measures to ensure the security of their systems.
The largest cloud computing service provider in the world indicated on Wednesday that the company had discovered that hackers had attempted to use the loopholes to launch attacks four days earlier.
The Amazon added that these loopholes “are now being widely used in the real-life web environment”.
Earlier that day, the United States National Security Agency (NSA) and British authorities warned that those loopholes placed websites around the globe at high risk of hacking. It has been reported that some European Governments have temporarily cut off some of the websites while the two security loopholes are being repaired.
These loopholes remained undetected for more than a decade until researchers identified them using the artificial intelligence models of Open AI and Google. The scope of the potential impact of loopholes highlights the ability of artificial intelligence to subvert network security patterns.
Earlier this year, Anthropic and Open AI successively launched new models with strong cybersecurity capabilities, triggering warnings from politicians and financial regulators, and prompting the United States Government to temporarily restrict the launch of these models.
On Wednesday, the United States National Security Agency warned all agencies using programming language called PHP to update the system immediately, otherwise the system should be offline.
The developers responsible for the maintenance of PHP had updated the publish software earlier that day to repair those loopholes.
According to the United States National Security Agency: “These loopholes often provide direct access to server document systems or databases. If it is not possible to complete the update immediately, it is recommended that the relevant agencies temporarily take the system offline until the loophole is completely repaired.”
Researchers from two United States cyber security agencies, Calif and Wiz, have discovered these loopholes. Wiz under the banner of the Google parent company Alphabet estimates that the flawed PHP programming language supports about 70 per cent of the global website operation.
On Wednesday, Open AI said: “We consider this to be one of the most significant cyber-security problems affecting us in recent years, given the extent of these loopholes and the widespread use of PHP. As far as we know, front-line artificial intelligence models pre-discover these loopholes by the attackers.”
Open AI adds: “This highlights the ability of artificial intelligence to detect security loopholes that may affect the Internet as a whole, but the same capability may also assist the attackers in rapidly developing attack tools and loophole utilization procedures.”
Open AI worked with PHP, Calif and Wiz to verify these gaps and develop the corresponding restoration patches.
According to the White House, federal agencies are assessing the possible impact of these loopholes on government systems and critical infrastructure.
The widespread use of PHP on the Internet is largely due to the fact that WordPress, the most widely used website-building platform worldwide, is mainly PHP-based and operates in this language.
The National Cyber Security Centre (NCSC) of the United Kingdom indicated that these loopholes could allow hackers to control web servers without passwords or other access documents.
The agency stated: “The British National Cybersecurity Centre is urging British agencies to take immediate action.”
Auto-fetched and translated by AI Atlas. All rights belong to the original author.