Cybersecurity researchers have published information about a now-fixed security flaw in the widely used figma-developer-mcp Model Context Protocol (MCP) server, which could have allowed attackers to execute arbitrary code remotely. Cataloged as CVE-2025-53967 with a CVSS score of 7.5, the issue involves a command injection vulnerability caused by improper handling of user inputs. A vulnerability in src/utils/fetch-with-retry.ts allows remote code execution (RCE) by falling back to child_process.exec with interpolated curl commands when fetch fails, enabling attackers on the same network or via DNS rebinding to inject malicious shell commands through crafted inputs. Imperva identified a design flaw in the Framelink Figma MCP server, which integrates AI assistants like Cursor, allowing attackers to exploit indirect prompt injections for unauthorized operations and potentially achieve full remote code execution, jeopardizing sensitive developer data. According to security researcher Yohann Sillam, the issue occurs when building command-line requests for the Figma API endpoint. The MCP client initiates contact with the MCP endpoint, receiving an mcp-session-id for further communications. A vulnerability in src/utils/fetch-with-retry.ts allows remote code execution (RCE) by falling back to child_process.exec with interpolated curl commands when fetch fails, enabling attackers on the same network or via DNS rebinding to inject malicious shell commands through crafted inputs. The flaw was patched in figma-developer-mcp version 0.6.3, released on September 29, 2025. As a preventive measure, developers are advised to avoid child_process.exec with untrusted input and instead use child_process.execFile, which prevents shell interpretation. Imperva and FireTail warned that as AI-assisted development grows, evolving security practices are critical highlighted by a flaw in Google's Gemini AI chatbot that uses ASCII smuggling to bypass filters, potentially enabling identity spoofing and data poisoning in enterprise environments, with similar vulnerabilities affecting other LLMs like DeepSeek and Grok.
GitLab has issued a critical security update to address multiple vulnerabilities affecting both Community Edition (CE) and Enterprise Edition (EE). The latest patch fixes four flaw...
Cybersecurity researchers discovered a very advanced form of the ClickFix attack method, one that now includes browser cache smuggling to covertly drop malicious files onto target ...
Security researcher Norbert Szetei published a final analysis and a working proof?of?concept for CVE?2025?37947, demonstrating a reliable path from an ordinary SMB user to local ro...