Docker has published a guide to five security risks that arise when AI agents are allowed to browse, run code and call tools ...