Azure Content Safety offers guardrails in the age of GenAI

Azure AI Content Safety, an AI-powered platform to help organisations create safer online environments, is now generally available. The service uses advanced language and vision models to filter out both user and AI-generated content that is deemed unsafe, including adult content, gore, violence and hate speech. The ease of access to such content has been an ongoing concern in the Asia Pacific region, where social media and messaging apps are popular.

Azure AI Content Safety offers:

- Multilingual proficiency, ensuring that users are safe online regardless of the language a particular piece of content is in.

- Severity indication metrics, which rates specific flagged content on a severity scale of 0-7. This allows users to swiftly assess a potential threat in a particular piece of flagged content and address it in a timely manner.

- Multicategory filtering, providing an at-a-glance summary of what kind of content has been deemed unsafe.

- Text and image detection, ensuring that all types of content, including those with visuals, are scanned.

Initially introduced as a part of Azure OpenAI Service, Azure AI Content Safety is now a standalone system. This means that customers can use it for AI-generated content from open-source models and other companies’ models as well as for user-generated content as part of their content systems, Microsoft said in a blog post.