Cato Networks Unveils AgentSense AI Security
The AI security tool detects prompt injection attacks and unsafe actions by AI agents before they execute.
Updated on Nov. 19, 2026 in Artificial Intelligence

Cato Networks introduced AgentSense, an AI security solution developed by Cato AI Labs to detect prompt injection attacks and unsafe actions by AI agents. The company also launched the AgentSense Challenge, offering $10,000 to each of the first three eligible participants who successfully complete it.
Why it matters
Enterprises deploying agentic AI face a growing security challenge from prompt injection and unsafe agent actions. AgentSense is designed to address that risk by monitoring how external content influences an agent's decisions.
AgentSense Untrusted Influence Detection traces how instructions and data influence an agent's behavior at runtime and blocks unsafe actions before execution. Cato has not disclosed detection rates or false-positive figures, and independent validation of these capabilities has not been reported.
The players
Cato Networks
Cato Networks is a network security company that offers cloud-based secure access and networking services to enterprises.
Shlomo Kramer
Shlomo Kramer is a co-founder of Check Point Software Technologies and serves as CEO and co-founder of Cato Networks.
Itay Ravio
Itay Ravio leads Cato AI Labs, the research group that developed AgentSense.
The details
AgentSense separates trusted instructions from untrusted content within the AI session and uses Drift Detection to identify when agent actions depart from the original task or intent. The AgentSense Challenge invites the security community to attack an AI agent workflow protected by the product.
Timeline
Cato Networks announced AgentSense on November 19, 2026.
The Tech Race
Prompt injection has been catalogued as a leading risk for LLM applications since OWASP began publishing its LLM Top 10, and AgentSense responds by tracing influence at runtime rather than only screening inputs. The product positions Cato against a growing field of vendors building security layers for autonomous AI agents.
Organizations rolling out AI agents that act on external content may gain a runtime control for blocking unsafe actions before they execute. The Challenge gives security researchers a public way to test the claims directly against a live agent workflow.
The takeaway
Teams building AI agents should treat untrusted content as a potential instruction source and verify agent actions against the original task. Independent testing of any such product remains essential before relying on it.
Further reading
For more on how AI systems are being secured as they gain autonomy, explore Artificial Intelligence coverage.
Source note: This article includes information reported by Channel Insider.



