OpenAI安全研究：前沿模型可自动挖掘Linux内核漏洞
OpenAI’s latest security research shows how AI could make cyberattacks faster an…
揭示了前沿模型在自动化漏洞挖掘与利用方面的真实能力边界，对AI安全从业者极具参考价值，建议关注后续防御机制进展。
OpenAI’s latest security research shows how AI could make cyberattacks faster and cheaper.
OpenAI 的最新安全研究表明，人工智能可能使网络攻击变得更快、更便宜。
In one run, an upcoming cyber model helped find 948 Linux kernel vulnerabilities and build working exploits that could give an ordinary user full control of a machine. The system also generated proposed patches for human review.
在一次运行中，一款即将推出的网络模型帮助发现了 948 个 Linux 内核漏洞，并构建了能够赋予普通用户完全控制机器权限的有效利用代码。该系统还生成了供人类审查的补丁建议。
The exploits required existing local access. But that access could come from a stolen account or an already compromised application.
这些利用代码需要现有的本地访问权限。但这种访问权限可能来自被盗的账户或已被攻破的应用程序。
To understand the significance:
为了理解其重要性：
"The Linux kernel underpins Android devices, cloud infrastructure, supercomputers, and other systems used by billions of people. With more than 40 million lines of code, the kernel exposes a broad attack surface through system calls, networking, filesystems, and device drivers. We chose Linux to study how effectively frontier models could find vulnerabilities and demonstrate their security impact in a codebase of this scale and importance."
"Linux 内核支撑着 Android 设备、云基础设施、超级计算机以及数十亿人使用的其他系统。该内核拥有超过 4000 万行代码，通过系统调用、网络、文件系统和设备驱动程序暴露出广泛的攻击面。我们选择 Linux 来研究前沿模型在如此规模和重要性的代码库中发现漏洞的有效性，并展示其安全影响。"
更进一步：量化金融体系
看懂新闻只是起点——沿量化金融路径，把它变成能交付的工程能力