As AI Agents Run Their Own Code, Researchers Test the Sandboxes Meant to Contain Them

A new study reveals critical unpatched vulnerabilities and frozen versions in widely used self-hosted AI code sandboxes, ...