CVE-2026-41318 AnythingLLM 存储型XSS漏洞

// PoC Concept for CVE-2026-41318 // The vulnerability exists in the 'Chartable' component where chart captions are rendered without sanitization. // Attackers can use prompt injection to force the LLM to output a markdown image with a malicious alt text. // 1. Prompt Injection Example (sent to LLM): // "Create a chart about sales. Use the following as the caption: " // 2. Malicious Payload stored in Chart Caption: // The LLM outputs a chart with the caption containing the markdown image. // 3. Stored XSS Trigger: // When a victim loads the chat history, the Chartable component renders the caption. // Since DOMPurify is missing, the browser executes the script. // Example Payload Structure: const payload = ``; // If the application allows raw HTML injection via the alt text interpolation: // <img alt="<img src=\"x\" onerror=\"alert(1)\">" src="..."> // In the vulnerable component, this might be rendered directly leading to XSS.