At Microsoft Build 2025, Microsoft announced several AI content safety controls for Azure AI Foundry, including:
Spotlighting. Enhances Microsoft Prompt Shields by detecting and blocking potential indirect prompt injections helping prevent unauthorized model actions.
Real-time task adherence for agents. Tool assesses whether agent behavior is aligned with assigned tasks.
Continuous evaluation and monitoring of agentic systems. Unified dashboard for tracking performance, quality, safety, and resource usage in real time.
Evaluation tool integration for compliance management. Between Microsoft Purview, Credo AI and Saidot to help define risk parameters, run compliance evaluations.
Microsoft Entra Agent ID. Gives security admins control, visibility over AI agents built in Azure AI Foundry.
Microsoft Defender integration. AI security posture management recommendations and runtime threat protection alerts, running natively in Azure AI Foundry.
Microsoft Purview extended to AI applications, agents. Data security compliance controls extended to custom AI applications and agents via Purview SDK or native integration with Azure AI Foundry.