Company Performance Metrics
Refractal is an AI cybersecurity company that provides a runtime trust and security layer for autonomous AI agents. Its software intercepts agent tool calls and data actions, evaluates them against enterprise policies and regulatory requirements, and then blocks, rewrites, escalates, or records those actions. The platform implements the Autonomous
Action Runtime Management specification to generate tamper-evident compliance receipts and align runtime behavior with standards such as the EU AI Act, NIST AI RMF, ISO/IEC 42001, and SOC 2 Type II. Refractal supports security, risk, compliance, and product teams by turning internal compliance demands into structured, testable policies that can be enforced at the point of action. It integrates with models and copilots including Claude, OpenAI, and Microsoft Copilot, and with observability tools such as Langfuse, LangSmith, and OpenTelemetry-based stacks. The system also supports adversarial evaluations and red teaming of agent and model configurations to detect prompt injection, data exfiltration, and other failure modes.
