1,800+ MCP Servers Exposed: AI Security Risks & Zero Trust

Over 1,800 Model Context Protocol servers are openly exposed to public access without any authentication. This alarming discovery comes as companies rapidly connect AI agents to their internal tools. Security researchers found 1,862 such servers during an internet-wide scan. When they checked a sample of 119, every single one allowed unauthenticated access to lists of internal tools. This means organizations are broadcasting their AI capabilities without even a password. What's more, these aren't just test servers. Forensic analysis shows production systems with write access to critical assets like financial databases, social media accounts, and customer relationship management platforms. This means businesses have connected their most important operations to AI agents but left them unsecured. This oversight is already leading to real-world threats, like the EchoLeak exploit. This zero-click vulnerability, discovered in June 2025, allows attackers to embed malicious instructions in documents. When AI systems like Microsoft 365 Copilot process these documents, they execute the hidden instructions, sending sensitive data to attackers without any user action. The bottom line is that neglecting AI security can lead to critical data breaches and operational compromises.