近期披露的“PHP2Shell”严重安全漏洞暴露了长期存在的技术债务问题,该缺陷涉及 NASA、维基百科、白宫以及所有运行 WordPress 或 Drupal 的网站 1。这一底层 Bug 源于 1994 年创建的 PHP 语言中的内存溢出错误,早在 2024 年已被报告但未受到足够重视 1。网络安全公司 Calif 在 OpenAI 的赞助下,利用 AI 工具在两周内发现了如何利用该内存错误执行任意代码的方法 1。Tolmo 公司的 Matt Suiche 指出,这是为过去 40 年由人类编写的遗留软件偿还技术债务的时刻 1。
人工智能大幅降低了发现深层逻辑漏洞的门槛,使得针对老旧代码的攻击风险激增,同时也推动了防御技术的发展 1。哥伦比亚大学的 Suman Jana 表示,AI 擅长处理需要非常具体推理的痛苦工作,但并非完全取代专家 1。作为应对手段之一,Anthropic 于 10 月 22 日发布了一项新工具,支持由大型语言模型生成补丁并证明其正确性,最后由形式化工具进行检查 1。尽管技术不断演进,约翰霍普金斯大学的 Matthew Green 仍悲观预测,直到发生更严重的事件(可能在 2027 年初)才会引发大规模资金修复投入 1。
The recent discovery of the "PHP2Shell" security flaw has become a stark example of how artificial intelligence is forcing the tech industry to address decades of accumulated technical debt 1. The vulnerability traces back to a memory overflow error in the PHP programming language, originally created in 1994 1. Although this bug was reported as early as 2024, it failed to attract significant attention until cybersecurity firm Calif conducted an independent audit 1. With sponsorship from OpenAI, Calif utilized AI tools to identify a method for exploiting this underlying bug to execute arbitrary code within just two weeks 1.
The impact of this vulnerability is widespread, affecting critical institutions including NASA, Wikipedia, and the White House, along with all websites operating on WordPress or Drupal platforms 1. Matt Suiche of Tolmo characterized the event as a necessary moment for paying off the technical debt incurred by legacy software written by humans over the past 40 years 1. While experts like Suman Jana from Columbia University noted that AI excels at the painful, detailed reasoning required to find deep logic flaws, they emphasized that these tools are not yet complete replacements for human expertise 1.
In response to the growing threat landscape, defensive technologies are also evolving rapidly. Anthropic released a new tool on October 22 designed to generate patches using Large Language Models (LLMs) and verify their correctness through formal methods 1. Despite these advancements, Matthew Green from Johns Hopkins University offered a pessimistic outlook, predicting that large-scale financial investment in remediation might not occur until an even more catastrophic incident takes place, potentially as early as the beginning of 2027 1.
评论
还没有评论,欢迎留下第一条。