You can use SaaS Agent Security as an external threat detection system for Microsoft Copilot Studio. While Microsoft Copilot Studio includes its own built-in safeguards, SaaS Agent Security provides an independent security layer for additional oversight. If SaaS Agent Security detects malicious intent, it instructs Microsoft Copilot Studio to block the prompt request.

This external threat detection system also detects toxic prompts to prevent the generative model from taking action on the request. Toxic prompts are user inputs designed to provoke the AI into generating content that is harmful, biased, or discriminatory. Such prompts could trick the model into violating corporate safety guidelines, such as producing hate speech, harassing language, or instructions for dangerous activities. Microsoft Copilot Studio, after applying its own safeguards, sends prompts to SaaS Agent Security. If SaaS Agent Security determines that the prompt is toxic, it instructs Microsoft Copilot Studio to block the prompt request.