Mitigating Skeleton Key, a new type of generative AI jailbreak technique

Microsoft has identified a new AI jailbreak technique called Skeleton Key, which bypasses guardrails in generative AI models, allowing users to manipulate the model to produce unauthorized content or behaviors. This technique, discovered during testing on various AI models, poses a significant threat by enabling the model to ignore its intended guidelines. Microsoft has taken steps to address this issue by implementing Prompt Shields in Azure AI-managed models to detect and block such attacks. Additionally, software updates have been made to enhance guardrail bypass mitigation in Microsoft's AI offerings. To protect against Skeleton Key attacks, Microsoft recommends input filtering, system message prompt engineering, output filtering, and abuse monitoring. By integrating Azure AI with Microsoft Security tools, such as Microsoft Purview and Microsoft Defender for Cloud, security teams can detect and respond to threats like jailbreak attacks in AI systems. Microsoft encourages customers developing AI applications to consider these mitigation strategies and leverage Azure's built-in tools for model evaluation and monitoring to safeguard against evolving AI threats.