Refractal

Refractal is an AI security company building infrastructure to defend organisations against autonomous AI threats. Its work covers prompt injection, jailbreaks and data exfiltration across large language models and AI agents, as well as adversarial stress testing of AI deployments and broader evaluation of AI cyber risk.

The company's platform allows organisations to automatically stress test their AI systems, evaluate cyber risk in a controlled manner, and enforce policy across models, tools, data and modalities. It is designed as an overarching security layer spanning AI models that operate under differing security policies. Refractal also conducts frontier AI security research, which it combines with established cybersecurity methods in its approach to AI risk.

Refractal was founded on cybersecurity and AI expertise from MIT, Microsoft, NASA and the U.S. Navy. It works with governments and organisations that require high-assurance security, including regulated and critical sectors, and is backed by deep-tech and cybersecurity investors. Its approach aligns with emerging AI standards, among them the EU AI Act, NIST AI RMF, ISO/IEC 42001 and SOC 2 Type II.