Check Point researchers identified a critical remote code execution vulnerability in Cursor, a popular AI coding tool, allowing potential attackers to alter approved Model Context Protocol (MCP) configurations undetected. This flaw, termed “MCPoison,” enables malicious command execution without user consent when the configuration is modified. Fortunately, Cursor released version 1.3 on July 29, which addresses the issue by requiring user approval for any MCP changes. Users are urged to update to prevent unauthorized access each time they open Cursor. The vulnerability highlights significant trust model weaknesses in AI-assisted development environments, particularly concerning collaborative settings that rely on shared configurations. Check Point’s research emphasizes the need for improved validation processes for MCP configurations to avoid exploitation. As AI tools evolve, the security landscape will likely see more vulnerabilities, and Check Point plans to publish additional findings to enhance security across the AI development ecosystem. Ensure your Cursor environment is secure by keeping your software updated.
Source link
Share
Read more