Salesforce's Enterprise AI Harness, announced at Dreamforce, combines Data 360, MuleSoft and Agentforce so that agents inherit only the task-level policies their owners define, according to coverage of the event.
The design follows a principle that security teams have been pushing as agents multiply: give an agent the narrowest permissions needed for a task, and no more. With surveys showing that large shares of staff time go to reviewing agent outputs, governance built into the platform is a selling point as much as a safeguard.
Source: SiliconANGLE