Menlo Security Extends MARS to Secure AI Assistants and Coding Agents in 2026

Enterprise adoption of AI assistants and coding agents is outpacing the security controls built to govern them.

Employees increasingly rely on tools like Microsoft Copilot, Gemini in Chrome, and Claude Code to draft content, analyze data, and generate production code. Each interaction pushes sensitive data into large language models and exposes the business to new attack techniques such as prompt injection, tool abuse, and silent data exfiltration. Traditional web proxies, DLP, and endpoint tools were not designed to inspect AI prompts, model responses, or agent-to-tool traffic — creating a growing visibility and control gap for CISOs.

What was announced

Menlo Security has extended its MARS (Menlo AI Runtime Security) platform to secure enterprise AI assistants and coding agents, including Microsoft Copilot, Gemini in Chrome, and Anthropic Claude Code. The extension delivers runtime protection against prompt injection, data exfiltration, and unsafe agent behavior directly at the point of use in the browser and developer environment.

MARS applies real-time inspection to AI prompts and responses, enforces policy on which data can leave the organization, and blocks malicious instructions embedded in web content, documents, or repositories. Security teams gain a unified view of AI usage across sanctioned and unsanctioned tools, with granular controls per user, group, application, and data classification.

Why this matters

For CIOs, CISOs, IT directors, and procurement leaders, the risk profile of generative AI has shifted from experimentation to daily production use. Prompt injection is now a documented technique in the OWASP Top 10 for LLM applications, and coding agents that autonomously read repositories, execute commands, and call external tools can be manipulated to leak secrets or ship insecure code.

MARS closes this gap without blocking productivity. Instead of banning AI tools — a strategy that pushes usage into shadow channels — organizations can allow approved assistants, apply consistent DLP and policy controls, and audit every interaction. This is aligned with emerging regulatory expectations around AI transparency, data protection, and secure software supply chains.

Technical details

• Coverage: Microsoft Copilot, Gemini in Chrome, Anthropic Claude Code, and other major AI assistants and coding agents.
• Prompt injection defense: Runtime detection and blocking of malicious instructions in web pages, documents, and repository content.
• Data exfiltration control: Inspection of prompts and responses with DLP policies for PII, source code, secrets, and regulated data.
• Agent governance: Visibility and policy enforcement on tool calls, file access, and command execution by coding agents.
• Deployment: Cloud-delivered via the Menlo Secure Cloud Browser, no endpoint agent required for core coverage.
• Integration: Feeds telemetry into SIEM, SOC, and XDR workflows for unified incident response.

Softprom and Menlo Security

Softprom is the official distributor of Menlo Security. Enterprise customers can evaluate MARS for AI assistants and coding agents, plan pilots, and receive implementation support through Softprom and its certified partner network.

Securing AI is not about blocking innovation, it is about giving security teams the same visibility and control over AI traffic that they already have over web and email

This content was prepared as part of the Softprom DistriFlow project — an automated system for monitoring and adapting vendor news. Original source: original article.