AI agents are becoming more capable, but are their sandboxes keeping up? Researchers have disclosed SharedRoot, a sandbox escape affecting Anthropic's Claude Cowork that lets an AI agent chain a Linux

AI agents are becoming more capable, but are their sandboxes keeping up? Researchers have disclosed SharedRoot, a sandbox escape affecting Anthropic's Claude Cowork that lets an AI agent chain a Linux kernel privilege escalation (CVE-2026-46331) with a writable VirtioFS mount to access files across the host macOS system during local execution Read the full technical analysis here 👇 https:// thecy