VentureBeat
Follow
MCP stacks have a 92% exploit probability: How 10 plugins became enterprise security's biggest blind spot
The Model Context Protocol (MCP), initially praised for its rapid adoption as an AI integration standard, has become a major cybersecurity vulnerability due to its design. Pynt's research reveals a dangerous network effect: the more MCP plugins used, the exponentially greater the risk of exploitation. Frictionless connectivity, MCP's key strength, fostered a lack of security, with authentication initially optional and authorization updates arriving late. This has led to a growing attack surface where connections amplify vulnerabilities, creating a supply chain risk. Researchers have identified real-world exploits, including command execution and data exfiltration, highlighting the severity of the issue. The lack of built-in security allows attackers to potentially control AI assistants through compromised tools. A significant portion of deployed MCP servers are vulnerable, lacking proper authentication and authorization. A layered defense strategy, including gateways, semantic layers, and knowledge graphs, is crucial for mitigating risks. Security leaders are advised to implement gateways, layered security, regular audits, and limit plugin usage. Investing in AI-specific security is essential to address this emerging risk.
