Coming Soon: Anomity Integration with Claude Cowork
- Claude Cowork does its work inside a local virtual machine on the employee's computer. That VM keeps the agent contained, and it also hides the agent's activity from the security tools on the host.
- Anomity is building an integration that runs a lightweight Anomity sensor inside the Cowork VM, through Anthropic's EDR sensor program for Claude Desktop.
- The sensor is built to stay small and predictable: an unprivileged systemd service with no kernel modules, one outbound HTTPS destination, no self-updates, and packages signed with a non-exportable key that you roll out through your MDM.
- It extends the coverage Anomity already gives you across desktop agents, MCP servers, skills and browser extensions, so Cowork stops being the one agent you can't see.
- The integration will be available in private preview in the coming weeks. Request early access.
Employees now hand real work to AI agents on their own desktops. Those agents open files, run commands and call tools on the user's behalf, and much of that happens where security teams have no line of sight. Today we're announcing an upcoming Anomity integration with Claude Cowork on Claude Desktop, built on Anthropic's EDR sensor program for Cowork. It brings Anomity's visibility and policy to the place Cowork actually does its work.
Cowork goes beyond answering a prompt. It plans a task, then carries it out step by step: reading documents, running shell commands and code, and using connected tools. On the user's computer, that execution happens inside an isolated virtual machine. Isolation is a sound design choice, because it keeps the agent away from the host. It also means the security tooling already on the laptop sees the VM start and stop, and nothing in between. For a security leader, that is a blind spot exactly where an autonomous agent is most active.
Visibility inside the Cowork VM
The integration places a lightweight Anomity sensor inside the Cowork VM. Claude Desktop checks the package's signature against the approved vendor key before it installs anything, and the sensor then reports to the same Anomity console that already covers the rest of your fleet. That gives security teams visibility into what the agent does inside the VM, tied back to the rest of their Anomity view.
Visibility is only half of it. Anomity applies the same policy model it uses on endpoints, so Cowork activity lands in the 90-day audit trail and routes to your SIEM, Slack or Jira alongside everything else, instead of living in a separate tool.
Built to be trusted inside the VM
- Least privilege. The sensor runs as the systemd service
anomity-cowork-sensor.serviceunder its own unprivileged user, in a hardened unit, with no kernel modules. - One destination. It connects only over HTTPS to
app.anomity.ai. No other outbound connections, no third-party telemetry and no self-updates inside the VM. - Signed releases you control. Packages are signed with a non-exportable AWS KMS key, and new versions reach users only when you deploy them through your MDM.
- Scoped enrollment. A per-tenant enrollment token, which you can revoke from the Anomity console, gives each VM its own device key.
- Both Desktop builds. It is tested on Claude Desktop from Anthropic and on Claude Desktop connected to Amazon Bedrock or Google Cloud Vertex AI.
- Clean removal.
systemctl disable --now anomity-cowork-sensor.service, then/opt/anomity/uninstall.sh.
Securing every agent on the endpoint
Cowork is one agent among many. The Anomity Endpoint Sensor already inventories the desktop agents, coding CLIs, MCP servers, skills, extensions and secrets on Windows, macOS and Linux, and the Browser Sensor shows which AI services people use and with which account. On agents that expose a hook, such as Claude Code, Anomity can allow, deny or log an action before it runs. Bringing the sensor into the Cowork VM closes the gap that isolation opened, so every agent on the endpoint shows up in one view. For the wider picture, see how to build an AI agent inventory and securing computer-use and browser agents.
This integration builds on our existing work with Anthropic. In June 2026, Anomity was approved in the Anthropic Cyber Verification Program, the program through which Anthropic verifies cybersecurity organizations whose defensive work overlaps with dual-use technique. That verification governs what we may ask Claude models for our security research. The Cowork integration addresses a different layer: what agents actually do on your endpoints.
The integration will be available in private preview in the coming weeks. Anomity collects metadata only, redacts secret values on the endpoint, and is SOC 2 Type II attested. To see Anomity on your fleet today and get early access to Cowork coverage, request early access with a 30-minute demo.
Frequently asked questions
What is the Anomity integration with Claude Cowork?
It is an upcoming integration that runs a lightweight Anomity sensor inside the local virtual machine Claude Cowork uses on Claude Desktop, through Anthropic's EDR sensor program. Activity inside the VM reports to the Anomity console alongside the rest of your fleet.
Is it available today?
The integration is not available yet. It will be available in private preview in the coming weeks. Request early access.
Why can't my existing endpoint tools see Cowork?
On the user's computer, Cowork runs commands and code inside an isolated VM. A sensor on the host cannot see inside it, so the agent's actions are invisible unless a sensor runs in the VM itself.
Does it cover Cowork sessions in the cloud?
No. Cowork tasks that run on Anthropic's servers cannot host a third-party sensor. The integration covers only the local VM on the user's computer.
Where does the sensor send data?
Only to app.anomity.ai over HTTPS. It makes no other outbound connections, sends no third-party telemetry and does not update itself inside the VM.
How is the sensor deployed and removed?
Signed packages are deployed through your MDM, and Claude Desktop verifies the signature before installing. To remove it, disable anomity-cowork-sensor.service with systemctl and run /opt/anomity/uninstall.sh.




