New Startup Launches 'Firewall' to Guard Against Risky AI Agents

A new cybersecurity startup, AIR Security, has come out of stealth mode with $50 million in funding to tackle an emerging problem: the security risks posed by AI agents, skills, plugins and MCP (Model Context Protocol) servers. As businesses increasingly adopt AI tools that can act autonomously or connect to other software, these components can carry hidden dangers such as malicious instructions embedded in prompts, overly broad permissions, or vulnerabilities inherited from the software supply chain.

The company's product acts as a firewall specifically for AI agents, evaluating these tools before and during use to flag malicious behaviour, excessive access rights, or supply chain weaknesses. This reflects a growing recognition in the security industry that AI systems introduce new attack surfaces beyond traditional software, requiring purpose-built defences rather than relying solely on existing network or endpoint security tools.

For small and medium businesses, this development is a signal rather than an immediate action item — most SMBs won't be building custom AI agents themselves, but many are beginning to use third-party AI tools, plugins, and integrations in their daily operations. Understanding that these tools carry their own risk profile is an important first step in managing AI adoption safely.