CrowdStrike brings Falcon sensor inside Anthropic's Claude Cowork VM

A blind spot inside the agent's virtual machine

When Claude Cowork runs code locally, it does so inside an isolated virtual machine on the user's computer. CrowdStrike explains that, from the host's point of view, this VM is a single opaque process. The agent's activity inside it, including the processes it spawns, the packages it installs and the network connections it makes, is not visible to the host's endpoint security sensor.

The company also points to a data risk: because host folders can be mounted into the VM, sensitive data can sit within the agent's reach. That combination of limited visibility and accessible data is the problem the integration targets.

CrowdStrike announced the move at Fal.Con Europe 2026. According to the company, it was built in partnership with Anthropic in response to enterprise customer demand, and it extends Falcon protection to AI agents across their lifecycle, from agent creation through job execution.

How the detection is meant to work

CrowdStrike describes a scenario in which a user asks Cowork for a script to align slides to a brand standard. While completing the task, the agent installs the latest version of a common Python package. Neither the user nor the agent knows that an adversary has compromised the package at its source. Once installed, the package reads host data through a mounted folder and sends it to a known malicious endpoint.

In that case, CrowdStrike says, the Falcon sensor running inside the VM would detect it. The company calls this the type of malicious activity and potential data exfiltration the integration is built to catch as agents execute tasks.

What the integration delivers

CrowdStrike lists three main capabilities:

• Visibility and detection where agents execute: process, package and network activity inside the Cowork VM will no longer be a blind spot, so teams can detect and investigate threats with the same Falcon telemetry they use elsewhere.
• The same visibility on macOS and Windows: the VM runs the same Linux environment on both, so coverage is consistent regardless of the operating system.
• Deployment and control through the Falcon platform: admins turn the integration on through their existing device management, the same way they deploy the Falcon sensor today, and the Falcon Linux sensor deploys inside the Cowork VM.

From there, the security team controls it through the Falcon platform, which places Claude Cowork under the same security control plane as the rest of the environment, according to CrowdStrike.

What it means for enterprise security teams

For organizations that let employees use Claude Cowork, the integration is meant to close a gap between host-level endpoint tools and what happens inside the agent's sandbox. Security teams would see agent activity in the same telemetry and console they already use, rather than managing a separate tool.

The beta is limited to macOS for now. Windows support is expected later in November, per CrowdStrike. The company published the announcement alongside other AI security news on the same day, but the Cowork integration is the only one tied directly to Anthropic's product.