Microsoft have announced Azure Prompt Shields and enhanced Azure AI Content Safety capabilities, targeting prompt injection attacks identified by OWASP as the top threat facing large language models today. The unified API analyses inputs to LLM-based solutions to guard against direct and indirect threats that can circumvent safety measures, exfiltrate sensitive data, or trigger unintended system actions.
Prompt injection attacks involve malicious actors inputting deceptive prompts to provoke unintended responses from AI models. Direct attacks include jailbreak attempts where end users input malicious prompts to bypass security layers and extract sensitive information. Indirect attacks embed malicious prompts within external content like documents or emails, causing AI models to inadvertently execute embedded instructions when processing the content.
Prompt Shields operates in real-time as one of the first real-time capabilities made generally available, integrating seamlessly with Azure OpenAI content filters. The system leverages advanced machine learning algorithms and natural language processing to identify and mitigate potential threats in user prompts and third-party data.
Key capabilities include contextual awareness to discern prompt context and reduce false positives, and Spotlighting, announced at Microsoft Build 2025, which enhances detection of indirect prompt injection attacks by distinguishing between trusted and untrusted inputs.
AXA integrates Azure's content filtering technology with its own security layer for its Secure GPT solution, preventing prompt injection attacks using Prompt Shields. Wrtn Technologies uses Azure AI Content Safety to maintain compliance across its Korean-localised AI platform, with Chief Product Officer Dongjae "DJ" Lee noting: "It's not just about the security and privacy, but also safety. Through Azure, we can easily activate or deactivate content filters."
Azure AI Foundry offers comprehensive risk evaluations for content risks including hateful content, sexual content, violent content, and jailbreak vulnerabilities. Microsoft Defender integration surfaces AI security recommendations and runtime threat protection alerts within development environments, helping close gaps between security and engineering teams.
Microsoft positions itself as a leader in prompt injection attack mitigation, leveraging decades of research and AI product development experience. The real-time protection capabilities address enterprise concerns about AI security while enabling organisations to pursue AI transformation with confidence. Integration across Azure's AI stack provides comprehensive defence mechanisms for organisations deploying generative AI applications at scale.