Microsoft Beefs Up Defenses in Azure AI

Microsoft announced new Azure AI Studio features aimed at improving the safety and reliability of generative AI applications by mitigating prompt-injection/jailbreak attacks and reducing model hallucinations. The updates include Prompt Shields to detect and delimit untrusted inputs, groundedness detection to flag ungrounded outputs, a system message framework for defining model behavior, and preview-stage safety evaluations and risk monitoring to detect problematic inputs and vulnerabilities before deployment.