Entra Internet Access now blocks prompt injection at the network layer

🚨 The Signal: Entra Internet Access prompt-injection protection became generally available on 31 March 2026. It blocks malicious AI prompts across apps and agents with universal network-level policy. Combined with Conditional Access, it stops prompt-based attacks before they reach your Copilots and agents.

The Impact

Indirect prompt injection via web content or documents can hijack an agent; this control filters it at the network, with no per-app code changes.

• Entra admins: a new network-layer control against prompt injection.
• Security teams: agents and apps are protected without per-app code changes.
• Risk: indirect prompt injection via web content or documents can hijack agents.

The Action

1. License the Microsoft Entra Suite / Global Secure Access and enable Entra Internet Access.
2. Enrol users into the Global Secure Access client or browser-based access.
3. Create a Global Secure Access security profile that enables prompt-injection protection.
4. Extend the profile to Copilot Studio and local agents to filter risky file movement.
5. Pair it with Conditional Access requiring a compliant device and phishing-resistant MFA for AI apps.

Domain: Entra · Impact: high · Workload: Entra ID · Essential Eight: User Application Hardening