Your organization deploys an autonomous AI agent that queries multiple internal data repositories to generate executive reports. The development team requests broad read access "so the model can learn what's relevant." As the security architect, what is the MOST appropriate approach?
A. Grant read-only access to all repositories and log every query for review
B. Provision a non-human identity with least-privilege, task-scoped entitlements
C. Route all agent queries through a human-approved request workflow
D. Use the developer's service account credentials for traceability