OpenAI 发布新网络安全模型,够吗

初创公司 Charge Robotics 的首席技术官兼联合创始人 Max Justicz 利用一个 OpenAI 模型发现了一处严重漏洞,一旦被利用,黑客就可能控制运行 Ubuntu 的服务器。Ubuntu 是一款被广泛使用的、基于 Linux 的开源操作系统。
Justicz 出于爱好对网络安全颇有兴趣,此前曾在同一套代码库中发现过一个漏洞。他向 TechCrunch 表示,这次他是通过这家 AI 公司于 7 月推出的 GPT 5.6 Sol,并借助其 Daybreak Blue 计划发现该问题的——该计划让网络安全防御方能够使用强大的前沿模型。
“我其实就是把 ChatGPT 对准同一批代码,然后说‘我之前发现过这个漏洞,它几年前已被修复,请找出同样严重或更严重的漏洞’,结果它两个小时后就把这个漏洞交给了我,”Justicz 说。“它极其复杂绕人,人类很难找到,但它确实能用!”
他将该问题报告给了 Debian 项目,后者负责维护涉事软件,并发布了修复该漏洞的补丁。目前尚无已知的漏洞被利用的证据。但即便如此,他发现该漏洞之轻松也表明,在借助 AI 模型的情况下连新手黑客都能利用的潜在漏洞,我们所看到的不过是冰山一角。
“我只能想象,如果有更多时间、又能接触到限制更少的模型,人们能做到什么。未来几个月里,我们应当预期会出现一些极其离谱的漏洞被找到并加以利用,”Justicz 在社交媒体上讨论其发现时写道。“这对这里的大多数人来说可能显而易见,但目前只要拿一段尚未被审计过的代码或二进制文件,把一个强大的 AI 模型对准它,就能获得巨大的超额收益。”
OpenAI 似乎已经注意到了。今天,这家 AI 巨头宣布,一款专注于网络安全的模型 ChatGPT Cyber 将向网络安全项目 Daybreak Red 中最高层级的用户开放,该项目为攻击性网络安全研究人员提供漏洞搜寻与开发工具。昨天,OpenAI 表示,同一款模型已被用于扫描 Linux 内核中的漏洞并准备补丁,共发现 948 个潜在问题。OpenAI 还在网页浏览器 Firefox 中发现了 81 个“新型”漏洞。
“我们的目标是大幅扩展 Daybreak,把先进的安全研究带给各种规模的组织、独立研究人员和开源维护者,”这家前沿实验室在一份声明中表示。
OpenAI 坚持认为,在 AI 模型变得更擅长黑客攻击、也更容易获取之前,守护数字基础设施的人只有很短的时间窗口可以采取行动。然而对网络防御者来说,这感觉像是前沿实验室抛出了一大堆新工作,远超他们所能调动的资源。Justicz 表示,他个人在订阅和 Token 上花费了约 5000 美元,才找到几个漏洞利用方式。
除了 Daybreak 的模型访问权限外,OpenAI 还宣布了一项 10 亿美元的计划,用于支持该项目中从事电网或供水系统等关键基础设施工作的网络安全专业人员。
Daybreak 项目本身也招致批评。Justicz 指出,这家前沿实验室最近似乎“nerfed”(削弱)了 Daybreak Blue,使其在发现漏洞方面变得没那么有用。一方面,限制潜在恶意行为者的访问权限或许是好事——Justicz 说他自己就是为了加入该项目才刚上传了驾照照片。另一方面,网络安全专业人士表示,如果他们用不上前沿技术,而恶意行为者却能部署防护机制已被消融(abliterated)的开放权重模型,那他们就等于被绑住了一只手。
尽管如此,OpenAI 表示只有“数千名”网络安全专业人士参与 Daybreak 项目,而潜在黑客的数量要多得多。
目前尚不清楚前沿实验室是否在这一问题上投入了足够的资源。OpenAI 技术团队研究员 Eric Wallace 表示:“我们现在正将大量前沿算力用于发现并修补开源代码库中的漏洞,完全是公益性质的。”
但这次 Linux 漏洞搜寻耗资 120 万美元,明显低于该公司为寻找流体力学中一个著名问题的潜在解法所投入的约 600 万美元 Token 开销。
如果修补全球所依赖的数字基础设施的时间窗口很窄,那或许需要更大的紧迫感。
当你通过我们文章中的链接购买时,我们可能会赚取少量佣金。这不会影响我们的编辑独立性。
Tim Fernholz 是一名记者,撰写科技、金融与公共政策领域的报道。他长期跟踪报道私营航天产业的崛起,著有《Rocket Billionaires: Elon Musk, Jeff Bezos and the New Space Race》。此前,他曾在全球商业新闻网站 Quartz 担任资深记者十余年,职业生涯始于华盛顿特区的政治报道。如需联系 Tim 或核实其联系方式的真实性,可发送邮件至 [email protected],或通过 Signal 向 tim_fernholz.21 发送加密消息。
正在加载下一篇文章
译文由英文原文机器翻译,术语按行业惯例统一,可在文末展开原文核对。
查看英文原文(OpenAI released a new cybersecurity model. Is that enough?)
OpenAI released a new cybersecurity model. Is that enough? | TechCrunch
Image Credits:Samuel Boivin/NurPhoto / Getty Images
OpenAI released a new cybersecurity model. Is that enough?
11:06 AM PST · November 5, 2026
Max Justicz, the chief technology officer and co-founder of the startup Charge Robotics, used an OpenAI model to find a severe vulnerability that — if exploited — could allow hackers to take control of servers running Ubuntu, a widely used open source Linux-based operating system.
Justicz, who has a hobbyist’s interest in cybersecurity and previously found a vulnerability in the same codebase, told TechCrunch that he found this issue using GPT 5.6 Sol, released in July, through the AI maker’s Daybreak Blue, a program that gives cybersecurity defenders access to powerful frontier models.
“I literally just aimed ChatGPT at the same body of code and said ‘I found this bug previously that was fixed several years ago, please find bugs as bad or worse,’ and it came back in two hours with this,” Justicz said. “It’s extremely convoluted and hard for a human to find, but it works!”
He reported the issue to the Debian Project, which maintains the software in question and issued a patch to fix the vulnerability. There has been no known evidence of exploitation. Still, the ease with which he found it suggests that we’ve only seen the tip of the iceberg when it comes to potential exploits that even novice hackers can utilize with the help of AI models.
“I can only imagine what one could do with more time and access to less neutered models. We should expect some extremely crazy bugs to be found and exploited over the next few months,” Justicz wrote in a social media thread discussing his finding. “This might be obvious to most folks here, but there is currently huge alpha in taking some code or binary someone has not audited yet and just aiming a powerful AI model at it.”
OpenAI appears to have noticed. Today, the AI giant announced a new model focused on cybersecurity, ChatGPT Cyber, would be made available to the highest tier of users in the cybersecurity program, Daybreak Red, which allows offensive cybersecurity researchers tools for vulnerability hunting and development. Yesterday, OpenAI said that the same model had been used to scan and prepare patches for vulnerabilities in the Linux kernel, finding 948 potential issues. OpenAI also found 81 “novel” vulnerabilities in Firefox, the web browser.
“Our goal is to dramatically expand Daybreak, bringing advanced security research to organizations of every size, independent researchers, and open-source maintainers,” the frontier lab said in a statement.
OpenAI has insisted that there is a narrow window of time for the people who safeguard digital infrastructure to take action before AI models become even better at hacking and easier to access. For cyber defenders, however, it can feel like the frontier labs have unleashed a mountain of new work that outstrips their resources. Justicz said that he personally spent about $5,000 on subscriptions and tokens to find several exploits.
Besides Daybreak’s model access, OpenAI has also announced a $1 billion initiative to support cybersecurity professionals in the program who work on critical infrastructure like the electrical grid or water systems.
The Daybreak program has its own critiques. Justicz notes that the frontier lab appears to have recently “nerfed” Daybreak Blue to make it less useful at finding vulnerabilities. On one hand, it may be good to limit access from potential bad actors — Justicz said he just uploaded a photo of his driver’s license to join the program. On the other, cybersecurity professionals say they have one hand tied if they can’t use frontier technology while bad actors can deploy open-weight models with abliterated guardrails.
Still, OpenAI says just “thousands” of cybersecurity professionals are participating in the Daybreak program, versus a far larger number of potential hackers.
It’s unclear if frontier labs are devoting enough of their resources to this issue. Eric Wallace, a researcher on OpenAI’s technical staff, said that “we are now allocating a significant amount of our frontier compute towards discovering and patching vulnerabilities in open source codebases, fully pro bono.”
But the Linux bug hunt cost $1.2 million, which is significantly less than the approximate $6 million token spend the company invested in finding a potential solution to a famous problem of fluid dynamics.
If there is a narrow window of time to patch the digital infrastructure the world depends on, more urgency may be required.
When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.

Tim Fernholz is a journalist who writes about technology, finance and public policy. He has closely covered the rise of the private space industry and is the author of Rocket Billionaires: Elon Musk, Jeff Bezos and the New Space Race. Formerly, he was a senior reporter at Quartz, the global business news site, for more than a decade, and began his career as a political reporter in Washington, D.C. You can contact or verify outreach from Tim by emailing [email protected] or via an encrypted message to tim_fernholz.21 on Signal.
Loading the next article
Error loading the next article
原文为英文,正文由自动抓取并保留原文。
原文发布于 TechCrunch AI(OpenAI released a new cybersecurity model. Is that enough?),标题、摘要与正文由机器翻译,术语按行业惯例统一,未经人工逐句校订。