Attack Agent leverages large language models to systematically probe NLP applications for security weaknesses. It uses an agent-based workflow to autonomously craft adversarial inputs tailored to specific target APIs, execute these inputs, and parse responses to detect anomalies or unintended behaviors. Users can define custom attack modules, control the depth of fuzzing, and configure dynamic constraints. The tool supports batch processing of attack scenarios, automated reporting of discovered issues, and integration with CI/CD pipelines for continuous security validation. With extensible plugins and comprehensive analytics, Attack Agent empowers security researchers and developers to enhance the robustness and compliance of their AI-powered systems.