Embedded within the Authorisation Layer, Xybern Redact is a drop-in privacy proxy that strips PII from your prompts before they reach any AI provider, and restores the real data in every response. No model changes. No pipeline rewrites.
Xybern Redact sits between your application and any LLM provider. Before a request is sent, every piece of identifiable information is replaced with a consistent, fictionalized substitute. The model sees a complete, coherent prompt, just without the real data.
When the response comes back, Redact re-inserts the original values using a cryptographic vault. Your application receives a natural-language reply with the real names, numbers, and entities fully restored.
Point your existing LLM calls at the Xybern Redact endpoint instead of the provider directly. Redact handles anonymization before forwarding and de-anonymization before returning. Your application code stays unchanged.
Replace the provider base URL with the Redact endpoint. Every request through that client is now anonymized automatically.
The same entity always maps to the same pseudonym within a workspace. Multi-turn conversations stay coherent.
SSE responses are de-anonymized token by token. Latency is unchanged. Your streaming UI works without modification.
Redact works in two modes. When your team builds with LLMs, the API proxy handles it. When they use AI tools directly in the browser, the Chrome extension steps in, because there you have no API access at all.
Every anonymization is signed with an HMAC key and recorded in the Privacy Vault. Vault entries are chained into Merkle roots, giving you a tamper-evident log of every PII substitution, exportable as a signed audit trail at any time.
HMAC signing and Merkle chaining make it immediately detectable if any vault record is modified after the fact.
Seal a batch of vault entries into a Merkle root. Produce per-entry proofs on demand, without exposing the rest of the vault.
Export the vault as a signed PDF report. Suitable for GDPR Article 30 records of processing, HIPAA logs, and financial audit trails.
Redact policies define which entity types are intercepted, whether they are permanently redacted or restored in responses, and any custom regex patterns your organisation needs. Each workspace runs its own policy, independently controlled.
Some workflows should never see the real data re-inserted. Permanent redaction mode enforces a one-way anonymization with no restore path.
Add a regex for internal employee IDs, case reference numbers, policy codes, or any identifier your organisation uses. Redact treats them as first-class entities.
If PII escapes the proxy layer, a signed webhook fires immediately. Your security team is notified before the call completes.
The Xybern Redact Chrome extension brings PII detection and anonymization to any web page, no API integration required. Right-click selected text to anonymize it instantly, or paste any content into a field and let the extension detect and highlight identifiable data automatically.
ChatGPT, Claude.ai, internal tools, CRMs, email clients - the extension works on any page without needing an API integration.
Drag a DOCX into the extension popup and download an anonymized version in seconds. The vault records the full redaction event.
All extension anonymizations are vault-backed. The same audit trail, Merkle proof, and retention policy as the API proxy.
The Privacy Vault, cryptographic signing, Merkle integrity, and workspace isolation all run on the same infrastructure that underlies every Xybern product. One security posture across your entire AI stack.
Add Xybern Redact to an existing pipeline in minutes. Your team keeps using the models they already use.
