Robust Intelligence, now part of Cisco following its acquisition in 2024, is an AI security and validation platform that helps organizations protect their AI applications from adversarial attacks, data integrity issues, and model failures. Founded in 2019 by Yaron Singer, a professor of computer science at Harvard University, and Kojin Oshiba, and originally headquartered in San Francisco, California, the company developed technology to systematically test and secure AI models throughout their lifecycle. The platform's core product is the AI Firewall, which provides real-time protection for AI models in production by detecting and blocking adversarial inputs, prompt injections, data poisoning, and other attacks designed to manipulate model behavior. The AI Firewall inspects inputs to and outputs from AI models, applying validation rules and adversarial detection algorithms to prevent harmful or manipulated data from affecting model predictions. Robust Intelligence also provides automated AI testing through its Stress Testing product, which runs hundreds of configurable tests across categories including adversarial robustness, data integrity, bias and fairness, and model performance degradation. These tests can be integrated into CI/CD pipelines to validate models before deployment, acting as a quality gate for AI systems. The platform supports both traditional machine learning models and large language models, with specific capabilities for testing LLM applications including prompt injection detection, hallucination testing, and output safety validation. As part of Cisco's security portfolio, Robust Intelligence's technology is being integrated into Cisco's broader AI security and networking offerings. The platform is delivered as a SaaS solution and also supports on-premises deployment for organizations with strict data residency requirements. Pricing is enterprise-focused with custom contracts.

AI偏差检测

Robust Intelligence 将偏差和公平性测试纳入其 AI 验证平台的核心组件。其自动化测试框架评估模型的人口统计学偏差、不同影响和公平性违规情况，跨越受保护属性，帮助组织在 AI 模型部署到生产环境前识别和解决偏差问题。

AI内容审核

Robust Intelligence 的 AI Firewall 为语言模型提供输出验证，检测并过滤 AI 系统生成的有害、有毒或违反策略的内容。其实时检查功能帮助组织确保 AI 生成的输出在到达最终用户前符合安全策略和内容指南。

AI网络安全

Robust Intelligence 的 AI Firewall 为生产中的 AI 模型提供实时保护，检测和阻止对抗性输入、提示注入、数据投毒和其他旨在操纵模型行为的攻击。它检查 AI 系统的输入和输出，防止对抗性利用，充当专门为 AI 应用设计的安全层。

AI安全工具

Robust Intelligence 通过自动化压力测试提供全面的 AI 安全验证，评估模型在对抗鲁棒性、数据完整性、偏差和公平性方面的表现。其测试框架在部署前对 AI 模型运行数百个可配置测试，充当质量门，确保 AI 系统符合安全性和可靠性标准。

AI 测试工具

Robust Intelligence 通过其 Stress Testing 产品自动化 AI 模型测试，该产品运行涵盖对抗鲁棒性、数据完整性、偏差检测和性能降级的综合测试套件。这些测试集成到 CI/CD 管道中，使组织能够在部署前系统地验证模型并在开发期间捕获问题。