Claude-Assisted Research Finds 66 New TCP/IP Stack Vulnerabilities
Forescout's Vedere Labs used Anthropic's Claude and custom analysis tools to examine 10 embedded TCP/IP stacks, reporting 99 vulnerabilities in about 90 hours. The team said 66 were newly discovered, 52% were critical or high risk, and 54 might still affect the latest versions of the relevant software. Of 20 open-source and 34 proprietary software vulnerabilities reported to vendors or project maintainers, 19 and 5, respectively, had fixes when the report was published. The work took about 10 hours for AI-assisted static analysis and 80 hours for dynamic validation, with researchers still involved in verification. The team estimated AI-model costs of $1,200 to $2,300, excluding tool development and test-environment setup.
文/王珮瑤|2026-11-18 發表
網路安全業者Forescout旗下Vedere Labs近期利用現成AI模型,搭配自行開發的分析工具,針對10套嵌入式TCP/IP協定堆疊進行安全研究,在約90小時內找出99項漏洞,其中66項是這次研究新發現的漏洞,其餘33項則是過去已知的漏洞。這類協定堆疊廣泛應用於工控、醫療、網路及物聯網設備。
在66項新漏洞中,52%屬於重大或高風險,部分可能造成設備運作異常或阻斷服務。研究團隊認為,其中54項漏洞可能仍影響相關軟體的最新版本,因此將這些漏洞通報廠商或專案維護者,包括20項開放原始碼軟體漏洞及34項專有軟體漏洞。截至報告發布時,前者已有19項完成修補,後者則僅有5項完成修補。
這項研究延續2020年展開的Project Memoria計畫。當時研究人員主要依靠人工分析與傳統工具,程式碼分析約耗費9個月;如今研究團隊利用Anthropic的Claude模型搭配自建分析工具,約花10小時完成AI輔助靜態分析,再以約80小時進行動態驗證。雖然兩次研究的分析範圍不同,但成果顯示,AI有助加快漏洞挖掘與驗證速度。