As reported by the Financial Times, Upwind Security has introduced two new tools designed to protect the fast-growing world of enterprise AI agents.

The first, Upwind AI Detection and Response, is now generally available and monitors AI agents while they’re running in real time. Instead of only checking for problems before an AI system is deployed, the platform watches how each agent behaves in production. If an agent suddenly accesses unusual data, connects to an unexpected system, or acts outside its normal pattern, security teams are immediately alerted. The platform can also identify which agent was compromised, what data was touched, and recommend steps to contain the threat.

Upwind also unveiled its AI Agent Context Scanner, which reviews an AI agent’s instructions, tools, memory, and Model Context Protocol—or MCP—connections. The goal is to identify hidden or malicious instructions before they can manipulate an AI agent into taking unauthorized actions.

The announcement comes as more companies give AI agents access to sensitive business systems, customer data, and enterprise credentials. Cybersecurity experts say continuous monitoring is becoming essential as AI takes on more autonomous responsibilities.

This story was created with help from Omri Hurwitz Media.