隨著 DevOps 與雲端技術的普及，容器映像檔成為軟體供應鏈的 重要環節。然而，若映像中包含已知漏洞或惡意程式，將對系統安全 構成威脅。現有映像掃描工具如 Trivy [1] 和 Docker Hub [2] Scanner 雖可進行漏洞檢測，但依賴固定 metadata 與檔案結構，存在被繞過 的風險。本研究提出多種映像檔操控技術，包含刪除 metadata、重命 名檔案及結構混淆，成功繞過主流靜態掃描器，並於實驗中證實可隱 匿漏洞映像，揭示靜態掃描技術的防護侷限，並為未來防禦機制提供 改進依據。;In the context of the widespread adoption of DevOps and cloud environments, container technology provides strong isolation and flexibility, making software development and deployment more efficient. A Docker Image serves as the foundation for container execution, containing an application and all its dependencies, allowing users to quickly build and deploy applications. However, if an image contains security vulnerabilities, it may pose significant risks to automated deployment environments. For instance, attackers can exploit vulnerable dependencies (CVE) or inject malicious backdoors to launch attacks on the system, potentially leading to the leakage of confidential information. This highlights the inseparable relationship between container security and image integrity. Currently, many organizations rely on official or third-party image scanners to assess image security and mitigate potential risks. However, our research reveals that existing scanners still have limitations—attackers can craft specially designed images to bypass scanning mechanisms, thereby introducing vulnerable dependencies into containers and creating new attack vectors. This study aims to explore methods for bypassing current image scanner detection mechanisms and analyze their potential impact on containerized environments, with the goal of developing more effective defense strategies.
