Microsoft introduces 5 new AI tools to be integrated with Azure AI. -

Microsoft has rolled out new tools in Azure AI Studio to aid generative AI app developers in addressing quality and safety concerns linked with AI. These tools are either currently accessible or will soon assist developers in crafting high-quality and secure AI applications.

1. Prompt Shields

The integrity of AI systems is significantly jeopardized by prompt injection attacks, enabling malicious actors to manipulate AI to generate undesirable outcomes.

In response to this challenge, Microsoft has introduced Prompt Shields, an advanced solution that identifies and neutralizes both direct and indirect prompt injection attacks in real-time.

Direct prompt injections, known as jailbreak attacks, involve manipulating AI prompts to bypass safety measures, potentially resulting in data breaches or the creation of harmful content.

Microsoft’s Prompt Shield for jailbreak attacks, initially launched in November as ‘jailbreak risk detection,’ is tailored to detect and block these threats effectively.

Prompt Shield blocking a jailbreak attack

2. Groundedness Detection

Microsoft is introducing Groundedness detection, a feature aimed at identifying and rectifying ‘hallucinations’ in AI outputs—cases where the AI generates content that is ungrounded or misaligned with reality.

This tool is essential for upholding the quality and reliability of AI-generated content.

3. Safety System Messages

Microsoft is launching safety system message templates to bolster the reliability of AI systems. Developed by Microsoft Research, these templates steer AI behavior towards producing safe and responsible content, facilitating developers in building high-quality applications more efficiently.

4. Safety Evaluations

Acknowledging the difficulties in assessing vulnerabilities in AI applications, Microsoft is introducing automated evaluations for risk and safety metrics.

These evaluations gauge an application’s likelihood of producing harmful content and offer insights for implementing effective mitigation strategies.

5. Risks and Safety Monitoring

Additionally, Azure OpenAI Service now includes risk and safety monitoring for real-time tracking of user inputs and model outputs, enhancing AI deployment safety.

Finally, Microsoft is proud to introduce risk and safety monitoring within Azure OpenAI Service. This feature enables developers to monitor user inputs and model outputs for potential risks, providing insights to adjust content filters and application design for a safer AI experience.

These new tools from Microsoft Azure AI signify a significant step forward in creating secure and dependable generative AI applications.

By tackling key challenges in AI security and reliability, Microsoft remains at the forefront of responsible AI innovation, empowering customers to confidently scale their AI solutions.