AI agents are developing their own software supply chain.
Agents increasingly rely on tools, plug-ins, skills, MCP servers and other components that allow them to interact with websites, applications and external services.
Cybersecurity startup AIR has raised $50 million across two seed rounds to help companies monitor and secure those components. (TechCrunch)
The hidden infrastructure behind AI agents
An AI model by itself has limited ability to interact with the outside world.
Give it tools, however, and its capabilities expand dramatically.
An agent can potentially search information, interact with applications, access enterprise systems and perform multi-step workflows.
That makes the tools surrounding the model almost as important as the model itself.
AIR is building technology to identify agents operating inside companies and continuously evaluate the skills, tools and components those agents use. (TechCrunch)
Why conventional security tools may not be enough
Traditional software security often focuses on applications, endpoints and networks.
AI agents introduce another layer.
A company may not know that an internal agent is using a particular external tool or connecting to a third-party service.
The problem becomes even more complicated when agents can dynamically select tools based on what they are trying to accomplish.
The emergence of agent security
AIR’s approach is part of a broader market developing around what could become an AI-agent security stack.
That stack could eventually include:
- Agent discovery
- Identity management
- Tool verification
- Permission controls
- Activity monitoring
- Risk assessment
- Automated blocking
The objective is not necessarily to prevent companies from using agents.
Instead, the goal is to allow organizations to use them while maintaining visibility into what the agents can access and execute.
Security becomes part of AI deployment
As enterprises move from experimenting with chatbots to deploying autonomous agents, security infrastructure is becoming increasingly important.
AI agents may eventually become a normal part of enterprise computing.
If that happens, companies will need security systems designed specifically for software that can make decisions and take actions rather than simply display information.
AIR’s funding is another sign that startups are beginning to build that infrastructure.