radar AI Agents on Mac: ANSSI flags Claude Cowork and OpenClaw
ANSSI issues official warning about autonomous AI agents on workstations. Cowork Claude and OpenClaw named. Here's what you need to do.
ANSSI has just published an official alert about autonomous AI agents: OpenClaw and Claude Cowork are named, along with other open-source solutions. These tools run commands, read your files, send emails, and a malicious webpage could manipulate them without your knowledge. In other words: AI assistants on your Mac are also attack surfaces.
What you need to do
1. Disable access to sensitive apps (email, calendar, work files) for your AI agents in their respective settings.
2. Don’t use Claude Cowork, OpenClaw, or any autonomous agent on a machine containing client or financial data without proper supervision.
3. Check which applications each AI agent installed on your Mac has access to.
4. I’m preparing a complete guide to reduce risks: automatic filters, access segmentation, what you can disable without losing the tool’s usefulness: Coming soon in a tutorial
ANSSI issued an official warning on April 13, 2026, about autonomous AI agents on workstations. OpenClaw and Claude Cowork are specifically called out. These tools can run commands, read and write files, send emails - and a simple prompt injection can turn them against you.
I’m cooking up a comprehensive guide to reduce risks: automatic filters, access segmentation, what you can disable without losing functionality. Stay tuned.
Sources