字号·· | 护眼
纽约时报

人工智能帮助发现Java重大安全漏洞,该软件平台对互联网至关重要A.I. Helps Find Major Security Bug in Java, a Software Platform Vital to the Internet

点「原文对照」整页切到原文,或双击某段只看那段的原文。

安全公司Calif的发现表明,人工智能能够发现流行软件中存在的重大安全漏洞。Java漏洞指向了近乎所有财富500强公司所使用系统中的漏洞。

The discovery by security company Calif shows A.I.’s ability to find gaping security holes in popular software. The Java flaw pointed to vulnerabilities in systems used by nearly all Fortune 500 companies.

一家网络安全初创公司周二透露,它已使用一种新的人工智能模型,发现了Java编程语言中的一个关键软件漏洞。Java是一种流行的编程语言,被政府机构和大约90%的财富500强公司使用。

A cybersecurity start-up revealed on Tuesday it had used a new A.I. model to uncover a critical software bug in Java, a popular programming language that is used by government agencies and about 90 percent of Fortune 500 companies.

这是公司和政府面临的现实的最新证据,即人工智能正在彻底改变计算机漏洞发现领域,以如此之快的速度识别代码中的缺陷,以至于网络安全专业人员难以跟上步伐。

It was the latest evidence of a reality facing companies and the government, as A.I. is revolutionizing the field of discovery in computer vulnerabilities, identifying flaws in code so rapidly that cybersecurity professionals are struggling to keep up.

使用领先的人工智能模型,初创公司Calif在过去几个月中已发现了一系列严重漏洞。该初创公司最新的努力揭露了一个缺陷,使用的是OpenAI最近发布的GPT-6 Cyber模型,该缺陷存在于Java中。Java被银行、医院、航空业和科技巨头以及政府机构广泛使用,包括承担国家安全任务的机构。

Using leading A.I. models, the start-up Calif has now unearthed a string of serious vulnerabilities in recent months. The start-up’s latest effort exposed a flaw, using OpenAI’s recently released GPT-6 Cyber model, in Java, which is widely used by banks, hospitals, the aviation industry and technology giants, as well as government agencies, including those with national security missions.

“Java是现代经济的跳动心脏,”Calif高级研究员布鲁斯·邓说。“从实际意义上讲,所有主要公司”都将受到该漏洞的影响,他补充道。

“Java is the beating heart of the modern economy,” said Bruce Dang, a senior researcher at Calif. “Practically speaking, all major companies” will be affected by the bug, he added.

邓先生及其同事在过去几天里一直致力于与“五眼”情报共享联盟的所有政府机构以及私营部门的主要供应商进行协调。“五眼”联盟由美国、加拿大、澳大利亚、新西兰和英国组成,私营部门的主要供应商包括谷歌和微软。

Mr. Dang and his colleagues have worked in recent days to coordinate with all government agencies in the intelligence-sharing alliance known as Five Eyes — which consists of the United States, Canada, Australia, New Zealand and Britain — and major vendors in the private sector, including Google and Microsoft.

国家安全局没有回应置评请求。该局的机密网络周二因其描述为“例行软件维护”的原因而中断。目前不清楚该中断是否与处理Java漏洞有关。

The National Security Agency, whose classified network went down on Tuesday for what it described as “routine software maintenance,” did not respond to a request for comment. It was not clear whether that disruption was related to addressing the Java bug.

甲骨文公司拥有Java商标并提供商业Java授权,未回应置评请求。该公司周二发布了一款软件补丁。

Oracle, the owner of the Java trademark and provider of commercial Java licenses, did not respond to a request for comment. The company released a software patch Tuesday.

来自Calif及其他利用人工智能工具的团体的漏洞公告,令网络安全界震惊,并颠覆了此前人们认为几个月前一小群程序员所能实现的成就。这些工作涉及在人工智能模型公开发布前与之合作,帮助网络防御者发现未被察觉的问题,以便在恶意黑客发现并造成破坏之前加以修复。

Announcements about exploits from Calif and other groups using A.I. tools have startled the cybersecurity world and upended what was considered achievable by a small band of coders just a few months ago. Those efforts, which involve working with A.I. models before they are publicly released, help alert network defenders to undetected problems in order to fix them before malicious hackers find them and wreak havoc.

但Java的发现可能是Calif迄今最严重的一次,可能使黑客获得对敏感内部系统的远程访问权限。这一发现紧随近日一系列看似独立的中断事件之后——微软产品故障、全国航班停飞以及银行中断——这些事件动摇了公众对日常技术可靠性的信心,并引发了人工智能可能是罪魁祸首的担忧。(到目前为止,每起事件的解释似乎都相对良性。)

But the Java discovery may be Calif’s most serious yet, one that could enable a hacker to gain remote access to sensitive internal systems. It comes on the heels of a cascade of seemingly independent disruptions in recent days — outages of Microsoft products, a nationwide grounding of flights and banking interruptions — that have shaken public confidence in the reliability of everyday technology and fanned fears that A.I. may be to blame. (So far, the explanations for each appear to be relatively benign.)

英国国家网络安全中心首席技术官奥利·怀特豪斯表示,人工智能辅助的漏洞挖掘和网络攻击构建,大大简化了识别系统弱点的过程。“我们已经看到这种趋势在一段时间内逐渐成形,人工智能被用于发现软件以及我们所有人都依赖的数字服务中越来越多的潜在漏洞。”Java漏洞是Calif在过去几周内借助OpenAI模型发现的第二个重大漏洞。

A.I.-assisted bug hunting and cyberattack building has made it much easier to identify weaknesses in systems, said Ollie Whitehouse, the chief technology officer of the National Cyber Security Center in Britain. “We have seen this taking shape for some time, with A.I. being used to discover a growing number of latent vulnerabilities across software and the digital services we all rely on.”The Java vulnerability is the second major bug Calif has unearthed with the aid of OpenAI models in the past few weeks.

该团体上个月发现了PHP编程语言的安全漏洞,据估计,该语言被互联网上70%的网站和应用程序使用。

It discovered last month security flaws in PHP, a programming language estimated to be used by 70 percent of websites and apps on the internet.

PHP漏洞的披露引发了全球对消费者和企业面临重大中断的担忧,各公司争相在黑客发动攻击前修补PHP。包括美国国家安全局在内的政府机构警告企业,如果无法立即部署补丁,应让网站下线。然而,中断最终未像一些人预测的那样严重,因为后来发现黑客不容易逆向工程攻击方法。

The PHP disclosure set off fears around the world about significant disruptions for consumers and businesses and a mad scramble by companies to patch PHP before hackers could pounce. Government agencies, including the N.S.A., warned businesses to take websites offline if they could not immediately deploy a patch. Disruptions ultimately proved less serious than some predicted, however, as it became clear hackers could not easily reverse-engineer the attack method.

但PHP漏洞主要涉及帮助网站运行的软件。相比之下,Java为现代数字世界的基础设施提供动力,包括云计算架构、身份管理工具、电子邮件服务等。

But the PHP flaws chiefly concerned software that helps websites function. In contrast, Java helps power the nuts and bolts of the modern digital world, including cloud computing architecture, identity management tools, email services and more.

“Java无处不在,”Calif首席执行官泰·杜昂(Thai Duong)表示。他指出,该Java漏洞不易被武器化为攻击,这让防御者有更多时间修补问题。

“Java is everywhere,” said Thai Duong, the chief executive of Calif. He noted that the Java bug was not easy to weaponize into an attack, giving defenders more time to patch the issues.

“但漏洞的影响尾巴很长,”他补充道,预测该漏洞可能在未来五到十年内威胁不更新系统的系统。

“But vulnerabilities have a very long tail,” he added, predicting the bug could be a threat to systems that do not update in the next five to 10 years.

在最重要的发现中,Calif演示了如何利用该漏洞入侵Keycloak,这是一个广泛使用的身份验证和登录系统。Keycloak保护着数千个组织,包括CERN、惠普、联合健康保险和NASA。

Among the most significant findings, Calif demonstrated that it could use this bug to break into Keycloak, a widely used authentication and log-in system. Keycloak protects thousands of organizations, including CERN, Hewlett Packard, UnitedHealthcare and NASA.

来自Calif的安全研究员张光良(Quang Luong)在Java的base64解码器中发现了该漏洞。base64是一种常用算法,用于将二进制计算机数据转换为包含小写和大写字母、数字及标点符号的可读文本集。研究人员发现,在某些计算机上,当Java尝试解码64个“A”字符后跟一个等号时,会出现该漏洞:这个名为 Java4Shell 的漏洞,会通过篡改计算机内存的一部分,导致 Java 错误地执行代码,通常会使程序立即崩溃。

Quang Luong, a security researcher from Calif, discovered the bug in Java’s base64 decoder, a common algorithm that is used to translate binary computer data into a legible palette of text that includes lowercase and uppercase letters, numbers and punctuation. On certain computers, researchers found that Java encounters a bug when it tries to decode 64 “A” characters followed by one equal sign: AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA= This bug, called Java4Shell, causes Java to erroneously execute code by tampering with a section of computer memory, which typically makes the program immediately crash.

加州研究人员借助人工智能,证明他们可以通过精心安排内存——一种被称为“堆喷射”的技术——来破坏数据,从而获得对底层系统的完全未经授权的访问。

With the help of A.I., Calif researchers demonstrated that they could corrupt data by carefully arranging the memory — a technique called “heap grooming” — granting them full unauthorized access to the underlying system.

《纽约时报》在该初创公司分享了该攻击的概念验证细节后,独立验证了加州的发现。成功的测试耗时不到一小时。

The New York Times independently verified Calif’s discovery after the start-up shared proof-of-concept details about the attack. The successful test took less than an hour.

人工智能在网络安全领域的快速进步,呼应了国际象棋世界冠军加里·卡斯帕罗夫在近 30 年前输给 IBM 的深蓝电脑后所做的观察。

The rapid advances with artificial intelligence in cybersecurity efforts mirror what the chess world champion Garry Kasparov observed after he lost a match to IBM’s Deep Blue computer nearly 30 years ago.

他预测,最好的棋局不会由人类或电脑下出,而是由两者的混合体下出。他称这一概念为“半人马”,以神话中半人半马的生物命名。这一概念流行起来。多年来,半人马被视为最好的棋手,将棋艺推向新高度。

The best games of chess would not be played by humans or computers, he predicted, but by a hybrid of the two. The concept, which he called a “centaur” in a nod to the mythological half-human, half-horse creature, took off. For years, centaurs were seen as the best chess players, reaching new heights in the game.

如今,公司和政府正面临“半人马黑客”的威胁——半人工智能、半人类,且完全有能力在互联网上制造混乱。

Now companies and the government are contending with the threat of the “centaur hacker” — half A.I., half human, and fully capable of creating chaos on the internet.

目前,最令人震惊的发现仍需要人类专业知识。在过去几个月里,小型资深安全研究人员团队利用顶级人工智能模型,在几天或几小时内构建出曾经需要国家安全局等情报机构的精英黑客班花费数月才能完成的高级网络攻击。

The most alarming discoveries — for now, at least — still require human expertise. In recent months, small teams of veteran security researchers have used top-shelf A.I. models to build in days or hours the kinds of advanced cyberattacks that once took platoons of elite hackers at spy agencies like the National Security Agency months to construct.

OpenAI负责网络准备与安全的福阿德·马汀表示,加州大学伯克利分校的Java研究“展示了人工智能如何协助专家级安全研究人员发现我们所有人都依赖的软件中的严重缺陷”,并有助于推动协作式修复。“我们希望更多安全团队拥有这些能力和工具,帮助让关键软件对每个人都更安全,”马汀先生说。

Fouad Matin, who leads cyber preparedness and safety at OpenAI, said Calif’s Java research “shows how A.I. can assist expert security researchers with finding serious flaws in software we all depend on” and helps lead to collaborative fixes. “We want more security teams to have these capabilities and tools to help make critical software more secure for everyone,” Mr. Matin said.

谷歌信息安全工程师爱德华多·维拉·纳瓦表示,“一旦接到通知,我们就迅速在谷歌服务中推出了缓解措施和补丁,开发了检测机制以确保该漏洞未被利用,并向客户提供了指导。”安全专家预计,这些风险至少在短期内可能会增长。大约20年来,半人马国际象棋选手被广泛认为能够下出最好的棋。但最终,半人马被更先进的超级计算机所超越。

Eduardo Vela Nava, an information security engineer at Google, said that “once notified, we swiftly rolled out mitigations and patches across Google services, developed detection to ensure there was no exploitation of the vulnerability, and provided guidance to our customers.”Security experts expect those risks are likely to grow, at least in the short term. For about 20 years, centaur chess players were widely seen as capable of playing the best game of chess. But eventually, the centaurs were eclipsed by more advanced supercomputers.

杜昂先生认为,人工智能可能只需时间问题就能独立构建最先进的网络攻击,这可能开启一个不可预测问题的新时代。联邦调查局本周发布警报,警告称黑客正在使用新的自主人工智能黑客工具,针对全球各地的政府机构和企业。

Mr. Duong believes it may be only a matter of time until A.I. can independently build the most advanced cyberattacks, which could usher in a new age of unpredictable problems. The F.B.I. this week issued an alert warning that hackers were using new autonomous A.I. hacking tools to target government agencies and businesses around the world.

“我不知道我们还有多少时间——也许三个月,也许三年,”杜昂先生说。“但目前,最好的黑客是半人马黑客。”达斯汀·沃尔茨为《纽约时报》撰写网络安全和情报报道,驻华盛顿。

“I don’t know how much time we have — maybe three months, maybe three years,” Mr. Duong said. “But right now, the best hacker is the centaur hacker.”Dustin Volz writes about cybersecurity and intelligence for The Times. He is based in Washington.

迪伦·弗里德曼是《纽约时报》的人工智能项目编辑,调查一系列话题。他既有记者经验,也有机器学习工程师经验。

Dylan Freedman is the A.I. projects editor for The Times, investigating a range of topics. He has experience as both a reporter and a machine-learning engineer.

阅读媒体原文 ↗