Microsoft has rolled out two major security enhancements for Azure AI Content Safety and Azure OpenAI Service. The new tools, named Prompt Shields and Protected Material Detection, are designed to bolster security measures and safeguard intellectual property.
New Tool to Counter Prompt Injection Attacks
Prompt Shields is engineered to counteract both direct and indirect prompt injection attacks. The feature is already available in the AI Studio. Previously known as Jailbreak Risk Detection, direct prompt injection refers to instances where users manipulate vulnerabilities to prompt the language model to produce unapproved content. Indirect prompt injection involves embedding hidden commands in external texts, which can influence AI interactions.
Utilizing advanced algorithms and natural language processing, Prompt Shields detects and mitigates these risks. The tool works in tandem with Azure OpenAI Service content filters and Azure AI Content Safety, providing a robust defense mechanism for AI applications.
Enhanced Protection for Intellectual Property
The Protected Material Detection feature addresses concerns related to intellectual property violation in generative AI outputs. Debuting in a preview in November 2023, this feature examines language model outputs for similarities with an index of third-party content such as songs, articles, and other online material. It helps prevent potential copyright breaches by spotting possible infringements.
Users can indicate the specific content they want to monitor through API requests. The API analyzes the input and returns a Boolean value to signal the result. A ‘true' value indicates a detected infringement, while ‘false' denotes no matches.
Diverse Uses for Protected Material Detection
Protected Material Detection has a range of applications including creative writing platforms, automated social media content generation, AI-assisted journalism, and AI-driven recipe creators. On creative writing platforms, it helps avoid producing content that mirrors known copyrighted material.
For social media automation, it checks for compliance with copyright laws by scanning AI-generated posts. In news writing, it ensures that AI-produced drafts are free from inadvertent copyright violations. Similarly, for recipe generators, it verifies that AI-created recipes do not duplicate existing protected content.
Availability and Integration
Prompt Shields and Protected Material Detection are now accessible for use in AI applications through Azure AI Content Safety and Azure OpenAI Service. The features are aimed at improving the security resilience of AI systems and ensuring adherence to intellectual property regulations.
For additional information, users can refer to the Azure AI services blog available on Microsoft's official website.
