Microsoft launches new Azure AI tools to cut out LLM safety and reliability risks

Microsoft has introduced new tools for Azure AI to enhance the safety and reliability of generative AI applications. These include:

• Prompt Shields: A capability that uses advanced ML algorithms and natural language processing to analyze prompts and third-party data for malicious intent, blocking them from reaching the model.

• Prebuilt templates for safety-centric system messages: These allow developers to guide the model's behavior toward safe, responsible, and data-grounded outputs.

• Groundedness Detection: A feature that uses a fine-tuned, custom language model to detect hallucinations or inaccurate material in text outputs.

• Automated evaluations: These stress test the gen AI app for risk and safety, measuring the possibility of the app being jailbroken and producing inappropriate content, and providing natural language explanations to guide developers on mitigations.

Groundedness Detection is a new feature in Azure AI that works by using a fine-tuned, custom language model to detect hallucinations or inaccurate material in the text outputs produced by the model. This feature is designed to ensure the reliability of the content generated by generative AI applications.

Prompt injection is a type of attack where an attacker manipulates the input prompt of a generative AI model to bypass safety controls and generate personal or harmful content. Azure AI is addressing prompt injection through the introduction of Prompt Shields, which analyze prompts and third-party data for malicious intent and prevent them from reaching the model.

Microsoft's new Azure AI tools are currently in preview and are expected to become broadly available in the coming months. However, Microsoft has not provided a specific timeline for their general availability.

Microsoft's focus on safety and reliability in AI highlights the company's commitment to building trusted AI. By providing tools that enhance the security and dependability of generative AI applications, Microsoft aims to address enterprise concerns and attract more customers by ensuring high-quality, safe outputs from AI models.