Cato’s AgentSense Moves SASE Toward Controlling What AI Agents Do

Cato's AgentSense Moves SASE Toward Controlling What AI Agents Do

Cato Networks introduced AgentSense, which blocks AI agent actions steered by untrusted content, weeks after launching an endpoint client that runs alongside other vendors’ security stacks. The moves extend Cato’s AI security past the network, though agents running inside SaaS platforms remain a visibility gap for the whole category.

What Is Covered in This Article:

  • AgentSense: Cato’s new runtime control blocks agent actions when untrusted content steers them, targeting indirect prompt injection.
  • Beyond the network: Cato AI Endpoint covers desktop AI apps and local agents that network inspection cannot see.
  • The visibility gap: Agents running inside SaaS and cloud platforms sit outside network and endpoint controls, a limit the whole category shares.
  • Peer moves: Netskope, Zscaler, Palo Alto Networks, Check Point, Cisco, HPE, and Microsoft are all moving on agent controls from different starting points.
  • Proof ahead: False-positive data, not just bypass counts, will decide whether buyers leave blocking switched on.

The News: On November 17, 2026, Cato Networks announced AgentSense, a runtime capability from Cato AI Labs that assesses AI agents as they work. Its first component, Untrusted Influence Detection, separates trusted instructions from untrusted content (emails, web pages, documents, tickets, or tool responses) inside an agent session, traces how that content influences the agent’s behavior and outputs, and blocks an action when untrusted content steers the agent toward an unsafe step. At launch, it enforces on local agents such as coding agents and desktop assistants, through three built-in rules. Cato says the approach targets indirect prompt injection without relying on static classifiers or known attack patterns. A second component, Drift Detection, will flag agent actions that depart from the task the agent was given.

AgentSense is integrated into the Cato platform and will be included at no additional charge in Cato AI Usage licenses. Cato also opened the AgentSense Challenge, which offers $10,000 each to the first three participants who redirect a protected agent workflow into malicious actions.

Two related moves preceded it. On October 27, Cato launched Cato AI Endpoint, a client that discovers and inspects AI use across browsers, desktop AI applications, and local agents (including their Model Context Protocol (MCP) servers and skills), and that can block unapproved tool calls in supported local agents. It runs alongside existing security service edge (SSE), virtual private network (VPN), and endpoint detection and response (EDR) tools, routing only AI application traffic to Cato. On October 6, Cato achieved the Amazon Web Services (AWS) Security Competency in AI Security (Agentic AI category), covering traffic to and from Amazon Bedrock and Amazon SageMaker.

All three build on Cato AI Security, launched in March 2026 with technology from Cato’s September 2025 acquisition of Aim Security.

Cato’s AgentSense Moves SASE Toward Controlling What AI Agents Do

Analyst Take: Cato Networks’ argument, in its own terms, is that an AI agent can hold valid credentials and still be talked into doing the wrong thing, and that the place to catch it is the same converged platform that already sees users, devices, data, and traffic. In roughly six weeks, the company has put three pieces behind that argument: a cloud competency for AI workloads on AWS, an endpoint client for desktop applications and local agents, and now AgentSense, which judges the action an agent is about to take rather than the access behind it.

Influence Tracing Goes After the Hard Part of Prompt Injection

Indirect prompt injection works because the agent is already authorized, so an instruction hidden in a ticket or a web page borrows the agent’s permissions instead of stealing anyone’s. Tracing how untrusted content shapes a specific action is, to us, a more durable approach in principle than chasing known attack strings, and including it in existing AI Usage licenses lowers the bar to trying it.

The endpoint client fills a different gap. Desktop AI applications can pin their connections to their own certificates, which keeps network inspection out, so a client on the device covers ground that inline inspection cannot. Cato also pitches it as a way in for organizations on other secure access service edge (SASE) providers (“no need to buy SASE first”), a door it can use to keep serving the standalone buyers it inherited with Aim Security, though another endpoint agent from a vendor a buyer does not otherwise use is a hard sell, and we expect most uptake to come from existing Cato customers. Our March 2026 SASE Signal pointed in this direction, expecting Cato’s points of presence, now equipped with graphics processing units (GPUs), to support guardrails for large language models and autonomous agents, and noting a pragmatic turn toward alliances.

Visibility Depends on Where the Agent Runs

In the Annual State of Security survey from ETR, part of Futurum (February 2026, N=483), SASE and SSE ranked last of nine layers respondents consider critical to securing generative AI activity, with 8% placing it in their top three; the endpoint client reads, to us, as a response to that perception.

The harder seam is the agent that runs on neither the endpoint nor the network path. Many enterprise agents live inside SaaS and cloud platforms (Salesforce, ServiceNow, and Microsoft are illustrative rather than exhaustive), acting inside the provider’s cloud, while influence tracing needs both the untrusted content and the resulting action in view. How much agent activity any network or endpoint control can observe is, we expect, the question that will decide how far this category reaches, for every vendor named below. Only 3% of respondents in the same survey (N=517) describe agents in broad production with consistent enforcement.

Peers Are Converging on Agent Action Control From Different Starting Points

The competitive question is where enforcement sits and how it is sold. Netskope, the closest comparison, now classifies agent actions before they run; Zscaler governs agent-to-agent and MCP traffic through a broker; Palo Alto Networks is building an agent gateway on its Portkey acquisition; Check Point offers agent guardrails built on Lakera; and Cisco extends AI Defense to runtime tool use. HPE governs AI usage through its hybrid mesh firewall with prompt-level inspection. Microsoft, at Ignite the same week, made Defender runtime protection for local agents generally available on Windows and is pulling agents from Salesforce, ServiceNow, and other platforms into Agent 365 through API connections.

Bypass Counts Are Half the Evidence

Cato cites results on AgentDojo, a public prompt-injection benchmark, where it says AgentSense cut attack success to 0.05%, against 0.81% for the next-best defense it tested, and the AgentSense Challenge will put the control in front of human attackers. The same benchmark also measures whether legitimate tasks still complete, and publishing that figure, along with production data, would answer the question buyers will ask next, since a runtime control that interrupts real work is likely to get switched to monitor-only.

Extending coverage to SaaS-hosted agents through API integrations (as Cato already does for Bedrock workloads) and dating Drift Detection would close much of the remaining distance between the claim and what a buyer can verify.

What Buyers and Cato Should Test First

In a proof of concept, measure what share of your agent activity the control can see, counting agents inside SaaS platforms and not just those on managed laptops.

At contract time, confirm which Cato license carries AgentSense and which agents it covers, since the announcement places it in AI Usage licenses and the launch blog in AI Security for Users, with enforcement on local agents first.

Cato’s hurdle is to show that influence tracing holds up on real enterprise workflows with a false-positive rate low enough that teams leave blocking switched on.

What to Watch:

  • Will Cato publish AgentSense Challenge results? Bypass findings released alongside false-positive data from real deployments would give buyers the first independent read on how influence tracing performs.
  • When does Drift Detection ship? Cato describes it in future tense with no date; its availability, and whether it also arrives in AI Usage licenses at no extra charge, will shape how complete the offer looks.
  • Will rivals match the no-extra-charge pricing? Watch whether Netskope, Zscaler, or others fold agent action controls into existing licenses rather than selling them as a separate line item.
  • How fast will Microsoft’s local agent controls spread? Defender runtime protection for local agents reached general availability on Windows at Ignite; uptake among Microsoft-standardized customers will test how much endpoint room third parties keep.
  • Does AI Endpoint sell outside Cato’s SASE base? Named customers running it next to another provider’s SSE, or an update on the migration path for former Aim Security customers, would show whether the standalone pitch lands.

For more information, read the full announcement from Cato Networks. Read the Cato AI Endpoint announcement.


Declaration of generative AI and AI-assisted technologies in the writing process: This content has been generated with the support of artificial intelligence technologies. Due to the fast pace of content creation and the continuous evolution of data and information, The Futurum Group and its analysts strive to ensure the accuracy and factual integrity of the information presented. However, the opinions and interpretations expressed in this content reflect those of the individual author/analyst. The Futurum Group makes no guarantees regarding the completeness, accuracy, or reliability of any information contained herein. Readers are encouraged to verify facts independently and consult relevant sources for further clarification.
Disclosure: Futurum is a research and advisory firm that engages or has engaged in research, analysis, and advisory services with many technology companies, including those mentioned in this article. The author does not hold any equity positions with any company mentioned in this article.
Analysis and opinions expressed herein are specific to the analyst individually and data and other information that might have been provided for validation, not those of Futurum as a whole.
Read the full Futurum Group Disclosure.

Author Information

Fernando Montenegro

Fernando Montenegro serves as the Vice President & Practice Lead for Cybersecurity & Resilience at The Futurum Group. In this role, he leads the development and execution of the Cybersecurity research agenda, working closely with the team to drive the practice's growth. His research focuses on addressing critical topics in modern cybersecurity. These include the multifaceted role of AI in cybersecurity, strategies for managing an ever-expanding attack surface, and the evolution of cybersecurity architectures toward more platform-oriented solutions.

Before joining The Futurum Group, Fernando held senior industry analyst roles at Omdia, S&P Global, and 451 Research. His career also includes diverse roles in customer support, security, IT operations, professional services, and sales engineering. He has worked with pioneering Internet Service Providers, established security vendors, and startups across North and South America.

Fernando holds a Bachelor’s degree in Computer Science from Universidade Federal do Rio Grande do Sul in Brazil and various industry certifications. Although he is originally from Brazil, he has been based in Toronto, Canada, for many years.

Related Insights
LM Config Closes the ITOps Visibility Gap for AI-Ready Infrastructure
November 19, 2026

LM Config Closes the ITOps Visibility Gap for AI-Ready Infrastructure

LogicMonitor's new LM Config solution closes critical visibility gaps in IT operations by natively integrating configuration management with infrastructure monitoring, enabling enterprises to deploy trustworthy agentic AI in their environments....
T-Rex Adds Former Border Patrol Chief to Sharpen Federal AI Credibility
November 19, 2026

T-Rex Adds Former Border Patrol Chief to Sharpen Federal AI Credibility

T-Rex Solutions bolsters federal credibility by appointing former U.S. Border Patrol Chief Michael Fisher to its National Security Board of Advisors, addressing security and compliance barriers to AI adoption in...
Sangoma Bets on Native AI Integration to Defend Its Installed Base
November 19, 2026

Sangoma Bets on Native AI Integration to Defend Its Installed Base

Sangoma deployed Sonella Voice AI across its UCaaS/CCaaS platform, reaching 100,000+ customers and 2.7 million seats with intelligent voice agents for enterprise workflows....
Mashfrog Plants Its Flag in Toronto to Chase Agentic AI Demand
November 19, 2026

Mashfrog Plants Its Flag in Toronto to Chase Agentic AI Demand

Mashfrog Group enters Canada through its Toronto office and ICCO Canada membership, positioning itself to serve enterprise demand for agentic AI solutions across wealth management, procurement, utilities, and hospitality....
Scale Moves Into Hospital Operations With Five Purpose-Built AI Agents
November 19, 2026

Scale Moves Into Hospital Operations With Five Purpose-Built AI Agents

Scale launched Hospital Operations AI agents on November 16, 2026, designed to eliminate administrative and clinical backlogs while addressing enterprise adoption barriers around reliability and data security....
Collate's Ontology Studio Targets the Semantic Gap Killing Agent ROI
November 19, 2026

Collate’s Ontology Studio Targets the Semantic Gap Killing Agent ROI

Collate's Ontology Studio bridges the semantic gap causing hallucinations in enterprise AI agents by mapping business concepts to data assets, competing against hyperscaler solutions in a rapidly growing market....

Book a Demo

Welcome

The vision behind everything in Futurum’s Custom Research practice is this: research should show you what is happening, what comes next, and what to do about it. It should be personal to each audience, easy for people to grasp, and structured so LLMs can reason over it accurately. And it should be fast and turnkey; you want answers now, not another project to carry for quarters.

Whether you are defining business, channel, or go-to-market strategy; evaluating vendors or justifying ROI; or commissioning research to fill an emerging market need, we have your back, with a program that answers your questions with the objectivity and credibility to drive real decisions.

To do it, we bring unmatched data to bear: Futurum research, surveys, and market projections; validated market feeds; ETR’s 15 years of insight from 10,000 technology decision-makers; G2’s buyer and user data; and what our analysts hear every day. Add leading primary collection, from AI-moderated voice interviews to surveys and analyst-led interviews, all turnkey, and every project comes out credible, nuanced, and actionable.

And we don’t just drop the results in your lap. For internal work, we provide analyst-led sessions, interactive dashboards, and a range of formats. For market-facing work, Futurum delivers turnkey activation and amplification that actually gets seen, by people and by LLMs, through our media and share of voice. This is research that moves decisions and markets.

We will meet you wherever you are, from a fast-turn brief to a multi-year program, and shape the work to your goals, timeline, and budget. The right program for your moment.

If any of this is useful, I would love to talk.

Benjamin Brown, VP Custom Research, Futurum Research

Benjamin Brown

VP, Custom Research · The Futurum Group

Newsletter Sign-up Form

Get important insights straight to your inbox, receive first looks at eBooks, exclusive event invitations, custom content, and more. We promise not to spam you or sell your name to anyone. You can always unsubscribe at any time.

All fields are required






Thank you, we received your request, a member of our team will be in contact with you.