Cybersecurity
Cato Networks Unveils AgentSense to Stop Harmful AI Agent Actions Before Execution

Key PointsAgentSense targets indirect prompt injection by tracing how untrusted content steers an…
View more
- AgentSense targets indirect prompt injection by tracing how untrusted content steers an agent's actions.
- Valid credentials do not guarantee safe behavior because an agent can be manipulated into harmful actions.
- Cato's $10,000 challenge invites researchers to test whether AgentSense resists adversarial workflow redirection.
Cato Networks has unveiled AgentSense, a runtime security capability designed to stop harmful AI-agent actions before they execute. Developed by Cato AI Labs, the technology traces how untrusted content influences an agent’s behavior, addressing a risk that ordinary access permissions do not fully resolve: an agent can have legitimate credentials and still be manipulated into doing the wrong thing.
Cato says AgentSense is integrated into its platform and will be included at no additional charge within Cato AI Usage licenses. The announcement pairs the technology with an adversarial testing challenge offering $10,000 each to the first three eligible participants who successfully redirect a protected workflow into malicious actions.
Valid Permissions Can Still Produce an Unsafe Action
An enterprise agent may need to read emails, search documents, consult websites, and call tools to complete a task. Each source expands the information it can use, but also creates an opportunity for outside text to influence its next step.
Indirect prompt injection exploits that boundary. Instead of issuing an instruction directly as the authorized user, an attacker hides one in material the agent encounters while working. A document, ticket, web page, or tool response can contain text that attempts to redirect the workflow, expose information, or trigger an unrelated action.
Checking whether an agent can access an application remains important. It answers a different question from whether a particular action serves the task the user authorized. For example, an agent may be permitted to read business records without being authorized to send them to a destination suggested by a retrieved document.
AgentSense’s proposition is to evaluate that behavior during execution, with an opportunity to intervene before a sensitive action is carried out.
Tracing Untrusted Influence Inside the AI Session
The announced Untrusted Influence Detection capability separates trusted instructions from untrusted content within an AI session and traces how the latter affects behavior and outputs. Cato says it blocks an action when it detects untrusted material steering the agent toward an unsafe step.
The distinction is important because an attack does not have to look like a familiar malicious prompt. Cato describes AgentSense as assessing influence rather than relying on static classifiers or a catalog of previously observed attack patterns. That is intended to address new forms of indirect prompt injection as well as recognizable ones.
The release describes the technology as patented, but does not provide a patent identifier, detailed implementation, or independent effectiveness benchmark. Its claim of protection against previously unseen attacks should therefore be understood as the product’s intended security capability, rather than a guarantee that every prompt-injection attempt will be stopped.
The operational question is whether a security layer has enough session and action context to distinguish using a document as evidence from treating that document as authority. AgentSense is aimed directly at that distinction.
Drift Detection Is a Separate Planned Capability
Cato also describes AgentSense Drift Detection in forward-looking terms. The planned feature will evaluate whether an agent’s actions continue to serve the original task as a workflow progresses.
That complements influence detection without addressing exactly the same problem. Untrusted Influence Detection asks whether outside content is steering an action. Drift Detection is intended to ask whether the action still belongs to the objective, even if the individual step is technically permitted.
A workflow can gradually accumulate steps that no longer advance its purpose. Detecting that departure requires maintaining a view of the authorized objective across the session, rather than judging every tool call in isolation. The release does not specify a separate availability date for Drift Detection.
How AgentSense Fits Cato’s Existing AI Security Platform
Cato’s agentic AI security offering already describes controls for discovering agents and Model Context Protocol servers, profiling tool calls, and enforcing policies on agent interactions. Its broader platform combines networking, access, and security context.
The company also separates security for employee AI use from protection for applications. Its AI Security for Applications offering describes runtime visibility, prompt and response inspection, and configurable protections for deployed AI applications. That provides context for AgentSense’s focus on the integrity of actions taken during a workflow.
In practice, controlling AI use involves several related decisions: which services an organization permits, which data may reach them, and what an agent may do with the information it encounters. AgentSense adds a behavioral claim to that stack—identifying when untrusted content is driving the action itself.
Prompt-Injection Research Shows Why the Boundary Matters
Cato’s EchoLeak research describes an attack chain in which externally supplied content could cause Microsoft 365 Copilot to expose information from its context. The significance for agent security is the transition from reading outside material to acting on instructions embedded in it.
Its CurXecute research examined a different outcome: prompt injection through external content could lead a coding agent to modify an MCP configuration and trigger command execution. Cato’s write-up says Cursor fixed that vulnerability in version 1.3. The example illustrates a failure mode; it should not be read as a claim that the patched issue remains open.
Both cases show why content provenance matters once an assistant can reach sensitive data or execution tools. An interaction that begins as information retrieval can become a security event when retrieved text changes the workflow’s control flow.
A Public Challenge to Test the Protection
The AgentSense Challenge invites the security community to plant instructions in a protected workflow and attempt to steer the agent away from the user’s intent. Cato says findings and participant feedback will inform further evaluation and improvement.
The first three eligible successful participants will each receive $10,000, according to the announcement. The supplied release does not include the final challenge URL or its detailed rules, so eligibility and testing boundaries will depend on the published program terms.
For enterprises expanding agent autonomy, AgentSense addresses a concrete need: preserving the connection between the task a user authorizes and the actions an AI system ultimately executes. The technology’s value will depend on how consistently it can enforce that boundary while allowing legitimate work to continue.












