Cato Networks has unveiled AgentSense, which is designed to monitor AI agents at runtime and block unsafe actions. Earlier, the vendor announced Cato AI Endpoint, a client for AI use beyond the browser.

Cato Networks unveiled AgentSense. Developed by Cato AI Labs, the technology continuously observes AI agents and is meant to safeguard the integrity of their autonomous decisions, actions and outcomes. On October 27, the company had already announced the Cato AI Endpoint client. Both products aim to secure AI in the enterprise but address different points.

AgentSense: runtime verification

According to Cato, zero trust is no longer sufficient if it is limited to user access. Organizations would need to verify at runtime how agents use their permissions: an agent can hold valid permissions and still carry out actions outside its scope. The company says AgentSense evaluates such actions as they occur and stops unsafe ones before they cause harm. CEO Shlomo Kramer points out that agents increasingly take over decisions that affect revenue and business continuity. More autonomy, he says, requires a solution that can intervene before actions cause damage.

“Untrusted Influence Detection” separates trusted instructions from untrusted content within the AI session. If such content leads an agent toward an unsafe action, the action is blocked before it is executed. Cato says it relies neither on static classifiers nor on known attack patterns and aims to stop previously unseen attacks as well, such as indirect prompt injection. The announcement offers no evidence for this. “Drift Detection” identifies when actions deviate from the original task and checks over the course of a workflow whether individually permissible steps still fit the goal.

As background, Cato cites Gartner figures: 71 percent of organizations expect to deploy agentic AI within twelve months, but only 21 percent have governance capabilities for it. By 2028, uncontrolled agents are forecast to account for 25 percent of security breaches. Cato names as the cause that agents cannot reliably distinguish trusted instructions from untrusted data. Itay Ravia, head of Cato AI Labs, points to DuneSlide, EchoLeak and CurXecute as examples of vulnerabilities exploited through prompt injection. Nithin Reddy of Dayforce says the company is looking forward to seeing how AgentSense performs; the quote describes an expectation, not practical experience.

In parallel, Cato is launching the AgentSense Challenge: the security community is invited to use injected instructions to lure a protected agent workflow into malicious actions. The first three successful participants will each receive US$10,000. According to Cato, AgentSense is included at no additional cost in the platform’s AI usage licenses.

Cato AI Endpoint: AI use beyond the browser

The client is designed to discover AI use in browsers, desktop applications and local agents, inspect prompts and responses, and enforce policy on the endpoint. Cato argues that browser extensions do not capture AI interactions outside the browser. According to the company, the client links user, endpoint, data, prompt, response, agent activity, tool call and policy decision, making risky agents, MCP servers and skills visible before they act. It operates alongside existing SSE, VPN and EDR tools; only AI application traffic runs through Cato, which is said to allow deployment in weeks rather than months. Chief Product Officer Ofir Agasi describes a gap in conventional point products, which do not see AI interactions on the host itself; Chris DePuy, co-founder of 650 Group, sees it as extending AI security to the endpoint. PlayPower CIO Alfred Sobrinho phrases the benefit conditionally: the solution could help provide an overview of AI use.

By Jakob Jung

Dr. Jakob Jung is Editor-in-Chief of Security Storage and Channel Germany. He has been working in IT journalism for more than 20 years. His career includes Computer Reseller News, Heise Resale, Informationweek, Techtarget (storage and data center) and ChannelBiz. He also freelances for numerous IT publications, including Computerwoche, Channelpartner, IT-Business, Storage-Insider and ZDnet. His main topics are channel, storage, security, data center, ERP and CRM.

Contact via Mail: jakob.jung@security-storage-und-channel-germany.de

Leave a Reply

Your email address will not be published. Required fields are marked *

WordPress Cookie Notice by Real Cookie Banner