Looking inside the AI workspace
Claude Cowork can run code in a virtual machine, a separate computing workspace inside a computer. Ordinary security monitoring on the main computer cannot see the activity inside that workspace. [2]
Anthropic documents a beta for Team and Enterprise administrators to run approved endpoint detection and response (EDR) software, workplace security monitoring, inside local Cowork. Cloud Cowork sessions are excluded. [1]
The sensor may see prompts, files, and Claude’s answers. It sends data to its security vendor and can communicate beyond Cowork’s network allowlist, the destinations Cowork itself is permitted to contact. [1]
Ask about your vendor’s actual timetable
CrowdStrike’s November 3 announcement says its integration is in beta on macOS, with Windows support expected later in November. [2]
Palo Alto Networks’ November 2 announcement puts its Cortex XDR integration in private preview in the coming months. That is a limited future test, not general availability. [3]
This is a catch-up briefing on current guidance and partner announcements. The help page’s update label does not establish its first release date. We have not installed a sensor or tested its protection.
Our take
Seeing an AI helper’s actions can make a workplace easier to supervise, but monitoring also deserves a privacy decision. A security product’s name on a compatibility list does not answer either question for your organization.
For example, before asking an AI helper to organize supplier spreadsheets, give your IT team a harmless sample task. Ask them to show which actions reach the security console and which parts of the task appear in its logs. That checks the work you actually intend to do.
Agree on who can read those logs, how long they stay, and which vendor receives them. Confirm the vendor’s current operating-system support before planning a deployment. A small, reviewed trial gives you better grounds for that decision than an announcement alone.