We want to build AI agents that use internal data to create external interfaces (i.e. customer service bots). How should we be thinking about permissioning? How do we prevent agents from leaking internal data sources via prompt engineering or simply a mistake? @strandbrown

→ View original post on X — @replit