Enhance AI security with Azure prompt shield and Azure AI content safety

Protect your AI system with a prompt shield. This is an integrated API that analyzes input to LLM-based solutions to prevent direct and indirect threats.

Strong defense against rapid injection attacks

The AI security environment is constantly changing, and rapid injection attacks are emerging as one of the most important threats to today's generation AI app builders. This occurs when enemies manipulate LLM inputs to change behavior or access fraudulent information. According to the Open Worldwide Application Security Project (OWASP), rapid injection is the highest threat facing LLMS today¹. Helps protect your AI systems from this new threat with the security of Azure AI content. It features a unified API that analyzes inputs to LLM-based solutions to prevent direct and indirect threats. These exploits include avoiding existing safety measures, removing sensitive data, or obtaining AI systems that perform unintended actions within the environment.

A quick spurt attack

In a rapid injection attack, the malicious actor enters a deceptive prompt to trigger an unintended or harmful reaction from the AI model. These attacks can be divided into two main categories: direct and indirect rapid injection attacks.

When end users enter malicious prompts designed to bypass the security layer and extract sensitive information, a direct, rapid injection attack, including jailbreak attempts. For example, an attacker may encourage an AI model to leak sensitive data such as social security numbers and private emails.
Indirect or cross-prompt injection attacks (XPIA) involve embedding malicious prompts in seemingly harmless external content, such as documents or emails. When AI models process this content, they can inadvertently execute embedded instructions and compromise the system.

Prompt Shields integrates seamlessly with Azure Openai content filters and is available for the security of Azure AI content. It protects against many types of rapid jet attacks, and new defenses are added regularly once new attack types are revealed. By leveraging advanced machine learning algorithms and natural language processing, prompt shield effectively identifies and mitigates potential threats of user prompts and third-party data. This cutting-edge feature supports the security and integrity of AI applications and helps protect your system from malicious attempts to manipulate and exploit.

The prompt shield features are as follows:

Contextual perception: A prompt shield can identify the context in which the prompt is issued and can provide an additional layer of security by understanding the intent behind the user's input. Also, contextual recognition can distinguish actual attacks from real user prompts, resulting in less false positives.
Spotlight: Microsoft Build 2025 has announced Spotlighting, a powerful new feature that enhances the ability of rapid shields to detect and block indirect rapid injection attacks. By distinguishing between untrusted and untrusted input, this innovation strengthens developers to ensure more generate AI applications for hostile prompts built into documents, emails, and web content.
Real-time response: Prompt Shield works with real time And it is one of the first real-time features that will become available in general. You can quickly identify and mitigate threats before compromising your AI model. This proactive approach minimizes the risk of data breaches and maintains system integrity.

An end-to-end approach

Risk and Safety Assessment: Azure AI Foundry provides risk and safety ratings to allow users to evaluate the output of content risk-generating AI applications: hatred and unfair content, sexual content, violent content, content related to self-harm, direct and indirect jailbreak vulnerabilities, protected material.
Red Teaming Agent: It enables automated scanning and adversarial investigations to identify large-scale known risks. Help your team move left by moving from reactive incident response to aggressive safety testing early in development. Safety assessments also support red teaming by generating adversarial datasets that enhance testing and accelerate problem detection.
Robust controls and guardrails: Prompt Shields is just one of the robust content filters in Azure AI Foundry. Azure AI Foundry offers many content filters to detect and mitigate risk and harm, providing rapid injection attacks, ungrounded power, protected materials and more.
Cloud Integration Defender: Microsoft Defender integrates directly with Azure AI Foundry to surface AI security attitude recommendations and runtime threat protection alerts within your development environment. This integration helps bridge the gap between security and engineering teams, allowing developers to proactively identify and mitigate AI risks, such as rapid injection attacks detected by prompt shields. Alerts are visible in the Risks and Alerts tab, allowing teams to reduce surface area risk and build safer AI applications from the start.

Customer Use Cases

AI Content Safety Prompt Shields offers many benefits. In addition to jailbreak, rapid injection, and defense against document attacks, it helps ensure that LLM works as designed by blocking prompts to explicitly circumvent rules and policies defined by the developer. The following use cases and customer testimony highlight the impact of these features:

AXA: Ensure reliability and security

AXA, a global leader in insurance, uses Azure Openai to power secure GPT solutions. By integrating Azure's content filtering technology and adding a unique layer of security, AXA helps prevent rapid injection attacks and ensure the reliability of your AI model. Secure GPT is based on the Foundry model Azure Openai, and utilizes a model that has already been fine-tuned using human feedback reinforcement learning. Additionally, AXA can also rely on Azure content filtering technology. This allows the company to use prompt shields to add its own layer of security to prevent model intrusion, ensuring an optimal level of reliability. These layers are updated regularly to maintain high levels of protection.

WRTN: Scaling safely with Azure AI content security

WRTN Technologies, a leading Korean company, relies on the safety of Azure AI content to maintain compliance and security across its products. At its heart, WRTN's flagship technology integrates the array of localized AI use cases and services by Korean users to their daily lives. The platform combines AI-powered search, chat capabilities and customizable template elements to allow users to seamlessly interact with “emotional companion” AI infusion agents. These AI agents have attractive, realistic personalities and interact with their creators. Vision is a highly interactive personal agent unique to you, your data, and your memory.

Because the product is highly customizable to a particular user, the built-in feature toggles between content filters and prompt shields is extremely advantageous, and WRTN can efficiently customize the security measures of various end-users. This allows developers to continue to be compliant, customizable and respond to Korean users while expanding their products.

“Not only security and privacy, but also safety, you can easily activate or deactivate content filters through Azure. There are plenty of features that will add to your product performance.”

Integrate Prompt Shield into your AI strategy

For IT decision makers, manufacturers looking to enhance security in their AI deployments are strategic orders to integrate Azure's rapid shielding. Luckily, enabling prompt shields is easy.

Azure prompt shielding and built-in AI security features provide an unparalleled level of protection for the AI model, allowing organizations to leverage the power of AI without compromising security. Microsoft is a leader in identifying and mitigating rapid injection attacks, using best practices developed in decades of research, policy, product engineering and learning from building large-scale AI products, allowing you to achieve AI transformations with confidence. By integrating these features into your AI strategy, you can protect your system from rapid injection attacks and help maintain user trust and trust.

Our commitment to trustworthy AI

Organizations across the industry use Azure AI Foundry and Microsoft 365 Copilot capabilities to drive growth, increase productivity and create value-added experiences.

We are committed to helping organizations use and build trustworthy AI. That means it's safe, private and secure. Trustworthy AI is only possible when combined with our commitments, such as safe future initiatives and responsible AI principles.