menlo security logo

New in Menlo Agent Runtime Security (MARS)

Say Yes to AI. Then Scale it Safely.

Your teams want to transform how their business operates with tools like Copilot, Gemini, Claude Code, and Claude Cowork. MARS lets them — running every AI session in an isolated cloud, stripping prompt injection, and masking sensitive data before the tool acts, all governed by the browser security platform secures your people.

Request a Live Demo

AI Showed Up Faster Than the Controls for It.

Your teams didn't wait for a policy. They're pasting sensitive data into assistants, pointing coding agents at internal systems, and letting sidebars read whatever's on the page. The security model you built assumes a person on the other end — someone who hesitates, questions a strange instruction, and can be trained. An AI tool does none of that. It reads a poisoned page or a booby-trapped file literally, and treats the instruction hidden inside it as a command.

Watch the 2-Minute Prompt-Injection Demo

That's not a reason to slow AI down. It's a reason to give it the same governance you already give your people.

85%+
of enterprise workflows now run in the browser
80%
of enterprise data sits in systems without modern APIs — agents reach it through the browser
15%
of organizations already run autonomous agents largely unsecured
400B
web sessions secured on the Menlo Cloud every year

One Runtime for Every Way Your Teams Use AI.

Browser Sidebar Agents (Copilot, Gemini)

The Risk
A poisoned page or hidden instruction turns the sidebar against you; sensitive data leaks out.
What MARS Does
Sanitizes the page bi-directionally, strips prompt injection, and masks sensitive data in real time.

Desktop AI Assistants (Claude Cowork)

The Risk
The assistant has browser, SaaS, email, and messaging access with the same reach as your employee.
What MARS Does
Controls what it can reach, sanitizes files and pages, and blocks exfiltration with DLP controls.

Coding Agents (Claude Code)

The Risk
Web-connected coding agents can be hijacked to exfiltrate source code and secrets.
What MARS Does
Restricts destinations, strips injection from pages and files, and keeps a full forensic trail.

Autonomous and MCP Agents

The Risk
Agents building business automations browse and scrape the open web, vulnerable to injection at every hop.
What MARS Does
Governs each session in a disposable cloud runtime with least-privilege access and blind authentication.

Files Pulled Into AI (SharePoint, OneDrive)

The Risk
Indirect prompt injection hides inside the documents agents ingest as context.
What MARS Does
Sanitizes files at rest, removes injection and malware, and masks private data before the tool reads it.

Your People, Same Platform

The Risk
Two workforces, two security models is unmanageable.
What MARS Does
The policies securing your human browser sessions now extend to your agents — no separate tooling.
DLP (Data Loss Prevention): Menlo data-loss-prevention controls, referenced above.
Request a Live Demo

Menlo Governs the Session, Because That's Where the AI Actually Acts.

A Disposable Cloud Runtime for Every AI Session

Web content executes in the Menlo Cloud and is thrown away after — so malicious code is absorbed there, never in your environment.

Prompt-Injection and Goal-Hijacking Defense

MARS separates instructions from data and strips hidden instructions, poisoned content, and steganographic payloads before the tool reads them. This is what the demo video shows live.

Blind Authentication

Credentials are injected only at the verified destination and are never visible to the agent — so a hijacked tool has no secret to leak.

Least-Privilege Governance With a Full Audit Trail

Scope what each tool can reach, mask what it shouldn't see, and record every action — with human-in-the-loop takeover to observe, assist, or stop any session.
MARS is built on Menlo's patented Isolation Core and Adaptive Clientless Rendering (ACR), with Content Disarm and Reconstruction (CDR) applied to every file. It covers the HTTP, browser, and LLM channels via cloud proxy and MCP server deployments.
Request a Live Demo
“

“The next billion users will be AI agents, not humans.”

— IDC Spotlight,
From Browsers as a Risk to Browsers as Security Control Planes
, March 2026 (sponsored by Menlo Security)
Frost & Sullivan — Global Leader, 2026
First Browser Security Platform to govern AI agents
IDC-validated approach to session-based security

Read IDC's Take on Securing the Next Billion Users.

IDC on why the browser is now the control plane for both humans and AI agents — and what a session-based security model requires. Free with a quick form.

Stack of IDC report pages titled 'From Browsers as a Risk to Browsers as Security Control Planes; Securing the Next Billion AI Agent Users.'