11-19 更新 实时
首页/ AI 大模型/ 正文

OpenAI GPT-6 Cyber 能入侵系统并提出修复方案

AI 大模型 来源:The Decoder 2026-11-06 04:36 0 阅读 机器翻译
OpenAI GPT-6 Cyber 能入侵系统并提出修复方案

Nano Banana Pro 由 THE DECODER 提供提示词

OpenAI 发布了 GPT-6 Cyber,这是一款专门训练用于发现安全漏洞、开发可用的漏洞利用程序并提出补丁建议的 AI 模型。

在针对性测试中,GPT-6 Cyber 在 Firefox 中发现了 81 个此前未知的严重漏洞,Mozilla 已完成修补。

在 Linux 内核中,该模型复现了 1,724 个漏洞,并为其中 948 个开发出可用的漏洞利用程序,是其上一代模型的五倍多。

OpenAI 推出了 GPT-6 Cyber,一款专门训练用于搜寻软件安全漏洞的 AI 模型。

OpenAI 表示,该模型能够发现缺陷、构建可用的漏洞利用程序并提出补丁建议。它还应当始终留在被划定的边界之内,而近几个月来 OpenAI 的一些模型在这一方面一直难以做到。

该公司称,它训练 GPT-6 Cyber 遵守其运行防护机制,所采用的手段之一是一个受 Hugging Face 事件启发而设置的蜜罐。在没有任何特殊安全指令的情况下,GPT-6 Cyber 在 99.93% 的情况下避免了越界,其上一代模型为 94.49%。启用安全指令后,该模型在 100% 的测试中都未越界。总体来看,OpenAI 在其自有的 Preparedness Framework 的网络安全类别中将 GPT-6 Cyber 评为“Critical”,与它给“基础”模型 GPT-6 Astra 的评级相同。

在 ExploitGym 这一基准测试上——该测试检验模型能否利用远程系统中的特定漏洞——据 OpenAI 称,GPT-6 Cyber 的得分为 48.5%,而其上一代 GPT-5.6 Cyber 为 32.1%。在一项针对在无法获取源代码的情况下发现 Web 应用漏洞的新测试中,OpenAI 报告的命中率为 40.3%,而未经网络安全训练的标准模型 GPT-6 Astra 为 32.0%。

ExploitGym 结果:GPT-6 Cyber 以 48.5% 领先,紧随其后的是 GPT-6 Astra(Daybreak Blue),为 42.4%。启用防护机制的 GPT-6 Astra 得分为 0.0%。| 图片来源:OpenAI

完整结果见系统卡。所有基准测试数字均来自 OpenAI 自身。截至目前,尚未有对该模型的独立测试公开发布。

GPT-6 Cyber 已在 Firefox 和 Linux 内核中发现真实漏洞

据 OpenAI 称,GPT-6 Cyber 已在真实世界的安全研究中产出成果。在一次定向行动中,该模型在 Firefox 中发现了 81 个此前未知的严重漏洞,尽管该浏览器此前已用更早的 AI 模型进行过测试。该模型还表明,30 个已知问题比此前认为的更为严重。Mozilla 接受了这些报告,并已发布修复程序。

据 OpenAI 称,在 Linux 内核中,GPT-6 Cyber 复现了 1,724 个安全漏洞,并为其中 948 个开发出可用的漏洞利用程序,是 GPT-5.6 Sol 在相同条件下完成的 166 个的五倍多。该公司表示,已向 Linux 维护者报告了 304 项关键发现,并附上了建议的修复方案。

针对 Linux 内核的这次行动,OpenAI 构建了一套专用的自动化系统,由侦察智能体和分析智能体组成。它们首先扫描了内核中约 5,034 处与安全相关的位置,约占整个代码库的 7%,随后筛选出 2,308 个候选目标,并在隔离的虚拟机中确认了 1,724 个漏洞。在此基础上,该系统又开发出 OpenAI 此前已报告的 948 个可用漏洞利用程序。据 OpenAI 称,整个运行过程消耗的算力成本为 1,213,430 美元,折算下来每个漏洞利用程序约 1,279 美元。作为对比,Pwn2Own 和 TyphoonPWN 等项目历来为同类发现支付的赏金在 20,000 至 70,000 美元之间。

OpenAI 表示,这些智能体至少使用了十种此前连外部内核安全专家都不知晓的新型攻击技术。按 OpenAI 的说法,模型既复现了已知的攻击模式,也找到了自己的方法。这一说法有多大分量,取决于你对“自己的方法”界定得有多严格。对于 AI 模型究竟能否真正做出新发现,还是只是把训练中见过的东西重新组合,研究人员仍存在分歧。

OpenAI 目前仅向部分选定客户开放访问

据 OpenAI 介绍,GPT-6 Cyber 最初通过 API 和 Codex,面向一小部分选定的 Daybreak Red 客户开放。之后访问权限将逐步扩展至所有 Daybreak Red 客户。与此同时,GPT-5.6 Cyber 正向所有 Daybreak Red 用户开放,而面向 GPT-6 Astra 的 Daybreak Blue alpha 访问权限也将扩展到更多组织和独立安全研究人员。

OpenAI 承认,随着模型能力不断增强,滥用企图也在增加。一些行为主体和团体已尝试将模型能力转移到自己的系统中。

OpenAI 列出了若干应对措施:核查模型的行为是否符合用户权限、针对新披露漏洞的定向监控、通过其所谓的 Private Safety Processing 进行自动化安全检查、对 Daybreak Red 实施基于 IP 的访问限制,以及建议在具备人工监督的隔离环境中运行该模型。

OpenAI 尤其重视对发电厂、水处理设施等关键基础设施的保护。为了让防御方更容易获得访问权限,该公司提到了其投入十亿美元的 "Daybreak for Frontline Defenders" 资助计划,以及一个名为 "Daybreak Red for Critical Infrastructure" 的新下属小组,用于授权测试。

OpenAI 谈到存在一个有限的窗口期,在此期间可以利用顶尖模型来保障软件安全。其目标是把这个 Daybreak 计划扩展到各种规模的组织、独立研究人员和开源维护者。至于这是否真能奏效,则是另一个问题。Epoch AI 的研究人员刚刚展示,开放的 GLM 模型本可以利用 Firefox 中已知但尚未修补的漏洞。

没有炒作的 AI 新闻——由人工精选

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(OpenAIs GPT-6 Cyber can both break into systems and propose how to fix them)

Matthias Bastian

Nano Banana Pro prompted by THE DECODER

OpenAI has unveiled GPT-6 Cyber, an AI model specifically trained to find security vulnerabilities, develop working exploits, and suggest patches.

In targeted tests, GPT-6 Cyber found 81 previously unknown critical vulnerabilities in Firefox, which Mozilla has already patched.

In the Linux kernel, the model reproduced 1,724 vulnerabilities and developed working exploits for 948 of them, more than five times as many as its predecessor.

OpenAI has introduced GPT-6 Cyber, an AI model trained specifically to hunt for security vulnerabilities in software.

OpenAI says the model can find flaws, build working exploits, and suggest patches. It's also supposed to stay within its assigned boundaries, something some OpenAI models have struggled with in recent months.

The firm says it trained GPT-6 Cyber to respect its operational guardrails, using, among other things, a honeypot inspired by the Hugging Face incident. Without any special safety instructions, GPT-6 Cyber avoided boundary violations in 99.93 percent of cases, up from 94.49 percent for its predecessor. With safety instructions enabled, the model stayed within bounds in 100 percent of tests. Overall, OpenAI rates GPT-6 Cyber as "Critical" in the cybersecurity category of its own Preparedness Framework, the same rating it gave to GPT-6 Astra, the "base" model.

On ExploitGym, a benchmark that tests whether a model can exploit a given vulnerability in a remote system, GPT-6 Cyber scores 48.5 percent according to OpenAI, compared to 32.1 percent for its predecessor GPT-5.6 Cyber. On a new test for finding vulnerabilities in web apps without access to source code, OpenAI reports a hit rate of 40.3 percent, versus 32.0 percent for GPT-6 Astra, the standard model without cyber training.

ExploitGym results: GPT-6 Cyber leads with 48.5 percent, followed by GPT-6 Astra (Daybreak Blue) at 42.4 percent. GPT-6 Astra with safeguards enabled scores 0.0 percent. | Image: OpenAI

Full results are available in the system card. All benchmark numbers come from OpenAI itself. No independent testing of the model has been published so far.

GPT-6 Cyber already found real vulnerabilities in Firefox and the Linux kernel

GPT-6 Cyber has already produced results in real-world security research, according to OpenAI. In a targeted campaign, the model found 81 previously unknown severe vulnerabilities in Firefox, even though the browser had already been tested with earlier AI models. The model also showed that 30 already-known issues were more severe than previously thought. Mozilla accepted the reports and has already shipped fixes.

In the Linux kernel, GPT-6 Cyber reproduced 1,724 security vulnerabilities and developed working exploits for 948 of them, more than five times the 166 that GPT-5.6 Sol managed under the same conditions, according to OpenAI. 304 critical findings were reported to Linux maintainers along with suggested fixes, the company says.

For the Linux kernel campaign, OpenAI built a dedicated automated system of scout and analyst agents. They first scanned roughly 5,034 security-relevant locations in the kernel, about seven percent of the total codebase, then filtered those down to 2,308 candidates and confirmed 1,724 vulnerabilities in isolated virtual machines. From there, the system developed the 948 working exploits that OpenAI had already reported. The entire run cost $1,213,430 in compute, according to OpenAI, which works out to $1,279 per exploit. For comparison, programs like Pwn2Own and TyphoonPWN have historically paid between $20,000 and $70,000 per comparable finding as a bounty.

The agents used at least ten novel attack techniques that were previously unknown even to external kernel security experts, OpenAI says. In OpenAI's telling, the models replicated known attack patterns but also found their own approaches. How much weight that claim carries depends on how narrowly you define "own approaches." Researchers still disagree about whether AI models can actually make new discoveries or just recombine what they've seen during training.

OpenAI limits access to selected customers for now

GPT-6 Cyber is initially available through the API and Codex for a select group of Daybreak Red customers, according to OpenAI. Access will be expanded gradually to all Daybreak Red customers. GPT-5.6 Cyber is being opened up to all Daybreak Red users at the same time, and alpha access to Daybreak Blue for GPT-6 Astra will be extended to more organizations and individual security researchers.

OpenAI acknowledges that as models grow more capable, abuse attempts are rising too. Some actors and groups have tried to transfer model capabilities into their own systems.

OpenAI lists several countermeasures: checks that verify whether a model's actions match the user's permissions, targeted monitoring for newly disclosed vulnerabilities, automated safety checks through what it calls Private Safety Processing, IP-based access restrictions for Daybreak Red, and the recommendation to run the model in isolated environments with human oversight.

OpenAI is paying special attention to protecting critical infrastructure like power plants and water treatment facilities. To make access easier for defenders, the company points to its billion-dollar "Daybreak for Frontline Defenders" funding program and a new subgroup called "Daybreak Red for Critical Infrastructure" for authorized testing.

OpenAI talks about a limited window of opportunity in which top-tier models could be used to secure software. The goal is to expand the Daybreak program to organizations of all sizes, independent researchers, and open-source maintainers. Whether that will actually work is another question. Researchers from Epoch AI just showed that the open GLM model could have exploited known but still-unpatched Firefox vulnerabilities.

AI News Without the Hype – Curated by Humans

原文为英文,正文由自动抓取并保留原文。

本文由采集器自动同步

原文发布于 The Decoder(OpenAIs GPT-6 Cyber can both break into systems and propose how to fix them),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。

阅读原文 ↗