Neo integrates with Claude Cowork to monitor AI actions in isolated VMs

Closing the AI visibility gap

Claude Cowork is a platform designed to handle complex, multi-step instructions to produce finished work. To protect sensitive data, Cowork runs its most critical operations inside isolated virtual machines, which are self-contained computing environments. While this isolation is a standard safety practice, it often creates a blind spot for corporate security teams.

Most traditional security tools can detect that a virtual machine exists, but they cannot see the specific activities happening inside it. This means security professionals might not know which files an AI agent is opening, what processes it is starting, or which external hosts it is contacting. As companies use more autonomous agents, these actions happen in a black box that is difficult to monitor or investigate after an incident occurs.

Neo, an AI security control platform, is addressing this by extending its monitoring capabilities into these isolated spaces. The new integration allows the platform to identify the semantic intent behind every action an agent performs. This provides security teams with a complete understanding of what an agent is doing and what specific risks those actions pose to the broader corporate environment.

How the Neo sensor works

The core of this update is a specialized sensor that monitors activity directly inside the Cowork virtual machine. Unlike standard Endpoint Detection and Response (EDR) tools, which might flag suspicious behavior without context, Neo provides AI security intelligence. It looks at the specific actions of the AI components to determine their purpose and potential impact.

Neo combines this real-time visibility with its Neoverse software intelligence engine and broader endpoint telemetry. By merging these data streams, the platform can provide a more comprehensive view of the environment. It does not just see a process running; it understands the context of the AI's behavior relative to the other components it interacts with.

A key technical feature of this integration is the ability to map every single action back to its origin. The system links Cowork activities to the specific device, user, and session that initiated the request. This creates a clear trail of evidence that allows security teams to conduct faster and more accurate investigations into how an AI agent arrived at a certain result.

Governance for autonomous software

The primary goal of this integration is to help organizations govern AI as it becomes more autonomous. As enterprise applications begin to perform tasks independently, security teams need a way to maintain policy control. Shlomi Salem, Chief Product Officer and Co-founder of Neo, stated, "Security teams can’t govern AI they can’t see."

The platform aims to ensure that adopting AI does not result in new visibility gaps for the company. By providing a way to see into the virtual machines, Neo allows teams to maintain the benefits of isolation while still having oversight. This balance is necessary for companies that want to use powerful agents for sensitive work without sacrificing their security posture.

Neo's broader platform is designed to provide inventory, intelligence, and attribution for various types of agentic software. This includes not only AI agents but also AI-enabled applications, browsers, and identities. The goal is to provide a unified way to manage risk as software becomes increasingly capable of acting on its own across the enterprise.

Availability and next steps

Neo is making this new integration available to all of its existing customers. This means organizations already using the Neo platform can begin incorporating Claude Cowork monitoring into their security workflows. The update is intended to help these customers manage the risks associated with agentic activity more effectively.

For those looking to learn more about the technical specifics of the integration, the company is hosting a webinar on November 30th. This session will likely cover how the sensor interacts with the Cowork environment and how to interpret the intelligence provided by the Neoverse engine.

The company was founded by veterans from SentinelOne and has received backing from several major venture firms, including Andreessen Horowitz and Bessemer Venture Partners. Their focus remains on providing the tools necessary for organizations to adopt AI securely while maintaining the visibility required to manage enterprise-wide risk.