AI
Authentication Controls and Environment-Level Access Rules
A Copilot Studio maker at a fictional electronics firm builds a laptop-info agent without enabling login. In the Shridhar authentication demo (04:02), an anonymous user skips product questions and extracts the internal product roadmap, which the agent returns without hesitation.
Admins respond by creating an environment group for internal workloads and publishing a rule requiring Entra ID authentication. The no-auth option grays out in the maker’s portal immediately. A second rule limits sharing to viewer-only, and a third blocks guest access by default for new environments, covering apps and flows alongside agents.
Network Isolation and Connector Policies to Block Data Exfiltration
Pradeep builds a customer-feedback agent that queries Azure AI Search over a public network and immediately returns customer phone numbers. In the Pradeep VNet and connector policy demo (12:32), the admin disables public access, creates a private endpoint, and links the environment to a VNet policy so all data travels that path only.
An IP firewall then restricts access to the San Francisco campus IP range, blocking Pradeep when he connects from Washington. Advanced connector policies at the environment-group level can block any connector, including ones that were previously exempt, cutting the remaining exfiltration paths.
Resilience Architecture: Availability Zones, Staged Rollouts, and Monitoring
Azure availability zones replicate data synchronously across at least three physical locations within a region, achieving an RTO under five minutes. When a traffic spike caused a cache issue, partitioning kept the problem in one region and prevented a global outage.
“However, resiliency is a partnership.” — Rashmi Mansur
Staged rollouts move changes through internal rings. A code regression caught in ring one left 98% of customer environments unaffected. After every incident, the team asks whether a monitoring gap exists, then builds category-wide coverage. A gateway crash linked to a dependent Azure service was detected within minutes and restored within 45 minutes using a safety switch.
Self-Service Disaster Recovery: Failover and Failback from PPAC
Self-service DR in Power Platform went generally available April 30. In the DR failover and failback demo (24:09), enabling DR takes three steps: select the environment, toggle Disaster Recovery, save. Two modes are available: emergency response for real outages and DR drill for compliance tests. Both trigger continuous data replication to a secondary region roughly 300 miles away.
“Failover typically takes less than five minutes.” — Rashmi Mansur
Failover runs through validate, prepare, run, and complete phases, switches DNS, and puts the environment into emergency mode in the secondary region. All actions log exact timestamps and step durations in the environment history, giving full auditability without opening a support ticket.
Notable Quotes
availability is no longer optional. Rashmi Mansur · ▶ 19:56
However, resiliency is a partnership. Rashmi Mansur · ▶ 21:46
Failover typically takes less than five minutes. Rashmi Mansur · ▶ 26:29
Key Takeaways
- Enforce Entra ID authentication at the environment-group level to prevent anonymous access to internal agents.
- Private endpoints, IP firewalls, and advanced connector policies each block a distinct exfiltration path.
- Self-service DR in Power Platform is generally available; failover takes under five minutes from PPAC.