Zirah AI

Zirah AI is an enterprise AI security platform. It protects enterprise AI by sanitizing prompts and enforcing document-level permissions inside secure hardware — so sensitive data can be used without compromising privacy or security.

Enterprise AI security is two problems, not one

Zirah tackles enterprise AI security as two problems: data leaving with every prompt, and internal copilots that surface content users were never meant to see. Different threats, different controls.

Two products, one engine

Zirah Gateway is an outbound forward proxy. It intercepts calls to OpenAI, Anthropic, and Gemini, tokenizes PII, PHI, and PCI with format-preserving redaction, forwards the request, then rehydrates the response — logging every transaction to your SIEM. Under 150ms of overhead, and it deploys inside your own VPC.

Zirah Sovereign RAG Shield is an inbound enclave. It runs your RAG pipeline inside an AMD SEV-SNP confidential VM and enforces document-level permissions at the vector retrieval layer — before any chunk reaches the model — producing cryptographically verifiable audit evidence.

The Ghost Key problem

Every RAG copilot has a vulnerability most vendors never name: SharePoint permissions protect a file, but not the vectors it is chunked into. A semantic search returns the most relevant chunks regardless of who had access, so an internal copilot can synthesize restricted content even when the RBAC looks correct. The model is not the leak; the retrieval layer is. Zirah enforces identity-scoped access at that layer, before content ever reaches the model's context.

Built on Dyanet technology — coming soon as QGuard

Zirah AI was built on Dyanet technology. That same technology will also be available directly from Dyanet as QGuard — enterprise AI security you can deploy and verify yourself. Talk to us about QGuard, or visit zirah.ai.