MCP Servers Can Leak Enterprise Secrets via Plaintext Configs and Over-Permissioned Access
Security researchers warn that AI agent integrations using the Model Context Protocol may expose sensitive data before security teams detect the server.
Key Takeaways
- MCP servers can expose enterprise secrets via plaintext configuration files, over-permissioned access, and prompt injection.
- The exposure can occur silently, often before security teams are aware the server is running.
- Organizations should restrict MCP server permissions, secure configuration files, and monitor for prompt injection attempts.
Related Security News
JadePuffer Agentic AI Attacks Target Azure Tenants, Destroy Cloud Resources
Security researchers have observed the JadePuffer ransomware operator conducting agent-driven attacks against Azure cloud tenants. The attacks involve reconnaissance, credential theft, and the destruction of core cloud components. Details regarding the specific use of agentic AI remain reported but unconfirmed.
Carbonato Botnet Leverages Hermes Agent AI Framework to Compromise Docker Hosts
Security researchers have identified a new botnet campaign, tracked as Carbonato, that repurposes the open-source Hermes Agent AI framework to compromise Docker hosts. The malware leverages exposed container endpoints to execute arbitrary commands through Telegram integration and harvests AI API keys and other sensitive credentials stored on compromised systems.
