OpenAI Model Reportedly Helped Find High-Risk APT Vulnerability
Charge Robotics CTO Max Justicz said he used a specialized OpenAI model to find CVE-2026-109518 in APT, prompting him to notify Canonical. Debian and Ubuntu have patched the authentication-bypass flaw, which could let an attacker controlling HTTP traffic replace a verified package and execute malicious code with root privileges under specific conditions. Canonical rated the vulnerability high risk, with a CVSS 4.0 score of 7.7.
The flaw affects widely used Linux package-management tooling, and available updates address a path that can bypass package verification and reach root-level code execution.
文/李建興|2026-11-05 發表
Debian與Ubuntu發布APT套件管理工具安全更新,修補編號**CVE-2026-109518**的驗證繞過漏洞。攻擊者若能控制系統與套件伺服器之間的HTTP傳輸流量,可在特定條件下替換已通過驗證的套件,誘使系統安裝惡意程式,以root權限執行任意程式碼。其他採用APT的Linux衍生發行版也可能受到影響。
APT是Debian、Ubuntu等Linux發行版用來下載、安裝及更新軟體的套件管理工具。正常情況下,APT會依據經數位簽章保護的套件庫資訊,檢查下載檔案的完整性,防止遭竄改的套件進入安裝程序。不過,此次漏洞讓攻擊者有機會在驗證完成後替換套件內容,繞過原有的安全檢查。
Debian指出,當HTTP重新導向使兩個不同的套件下載工作指向相同URI,APT可能將下載工作合併。若此時發生暫時性HTTP錯誤並觸發重新下載,APT對檔案硬連結(Hard Link)處理不當,可能導致已驗證的套件遭惡意內容覆寫。由於.deb套件的安裝指令碼會以root權限執行,攻擊者便可能藉此控制系統。
#OpenAI#Charge Robotics#Canonical#Debian#Ubuntu#APT#CVE-2026-109518