CATO Innovation Spotlight — October.

AI is now a part of everyone’s day, especially as AI agents are showing up everywhere in the enterprise. Let’s start with local agents—things like Claude Code, Cursor, Codex, and GitHub Copilot. These have spread way past engineering tools into powerful helpers that teams from marketing to finance to HR use daily.

And Cato helps teams put guardrails around them from the start. We can see the agents in use, identify risky configurations—whether that be MCP servers and tools—and actually control what happens in each session. And because hooks install before the agent does, teams can have those controls ready from the very first session.

And we don’t stop at the agent itself. Skills are sets of instructions that tell an agent what to do. And because employees and their agents can download these skills from anywhere on the internet, they can come with malicious instructions baked right in or simply instruct an agent to act out of scope. Cato shows you installed and active skills in your environment, understands exactly what those skills enable your employees and their agents to do, and flags any risky settings. Teams can then monitor or block risky skills by policy. That way, we’re protecting both the agent and the actions it can take.

And there’s more beyond the local agents we use: managed and no-code agents that are incredibly easy to create, which is great for innovation but can create a visibility problem. And didn’t you say that this is the kind of scenario teams run into all the time?

Exactly. One customer was telling us about their HR team wanting to make it easier for employees to get access to benefits information. So they built an agent within their HR benefits platform, which was already connected to internal HR databases, and shared it with the team. The intention was great, but that agent was handling sensitive employee information with no safeguards built in because that team doesn’t usually think about these things.

Definitely. And because they’re built on external systems, they’re also incredibly hard to keep track of. Now Cato brings those agents under governance. We discover managed agents across Amazon Bedrock, Microsoft Foundry, ChatGPT agents, and Copilot Studio. Teams can see who owns an agent, how it’s configured, and, on supported platforms, how it’s actually used.

And finally, let’s talk about how Cato makes it easy to protect the custom apps and agents your team is building, wherever they’re building them or however the data flows.

Whether teams run models through platforms like Google Cloud’s Vertex AI, OpenAI, Azure, or Bedrock, or route agent traffic through an AI gateway, Cato helps them apply the same controls without redesigning their environment.

Yes. We’ve expanded support to AI gateways, including LiteLLM, Kong, Portkey, and Azure API Management, all using one simple connection.

And that simple connection means protection is so much easier. Once connected, teams can apply AI policies with guards that control risks like sensitive data exposure and protect against AI attacks like jailbreaks and prompt injection. This means teams can enforce policy where their AI traffic is flowing without creating another complicated integration project.

So from local agents and skills to managed agents to the apps and the infrastructure behind them, Cato helps teams adopt AI with visibility and complete control.

And we’re not slowing down. Our teams are working on some incredible developments. So stay tuned for the next edition to learn more.