Vulcan Protect

Vulcan Protect is a runtime guardrail product for generative AI systems in production. It monitors and controls both inputs and outputs of AI applications to detect and block unsafe or high-risk content, prompt injection attempts, and sensitive data exposure. The product provides configurable threat categories covering data exfiltration, harmful or sensitive outputs, and other high-risk content, and allows custom threats and rules to be defined for specific use cases. Users can set independent detection thresholds for inputs and outputs and adjust sensitivity by threat type. It supports custom off-topic enforcement through natural language topic policies that define banned or allowed subjects, restricting requests outside an intended use case. Real-time monitoring capabilities detect disallowed and high-risk content across text, images, audio, and files, block prompt injections, and apply data loss prevention functions such as PII identification and masking. Vulcan Protect is described as model and infrastructure agnostic, deployable in cloud, on-premises, or hybrid environments, and includes multilingual detection intended to account for linguistic and cultural nuance rather than keyword or translation-based filtering. The vendor states the product incorporates threat insights from its Vulcan Attack red-teaming research and references alignment with frameworks such as OWASP Top 10 for LLMs and Agentic Apps, MITRE ATLAS, and NIST AI RMF 1.0.