PII & Regulatory Guardrails
Redact PII, and keep every AI answer within the rules you must follow
Shroud is the platform's single engine for detecting and redacting personal data. On the AI path it redacts sensitive data in flight, so a model provider only ever sees protected content; on upload and ingestion paths it scans documents and records as they arrive. Where a value is protected reversibly, Shroud also governs getting it back: who may look, and the record that they did.
Personal data reaches models through prompts, uploads, images, and ingested records, not only the chat box. Shroud detects it across all of them and protects it before it leaves, with operators from redaction to tokenization. Reversibly protected values come back only through a permission-gated, audited reveal, and AI answers can be checked against the regulatory rules you set.
- Forty-plus PII entity types detected across text, documents, and images
- Anonymization operators including redact, mask, hash, encrypt, and tokenize
- Policy-driven scoping by org, team, role, and user
- Coverage for chat messages, file uploads, image content, and device data
- Token vault, in-chat protection indicators, and a governed, audited reveal workflow
What you get
What Shroud gives your team.
Detection in text
Finds personal data with a configurable confidence threshold, entity filter and language, shipping English, Spanish, German and French language models.
Documents and images
Scans plain text, CSV, Markdown, PDF and Word files, redacts text in images with OCR, strips image metadata, and scrubs attachment filenames and sender names.
Seven protection operators
Replace, redact, mask, hash, believable synthetic fakes, deterministic tokenization or encryption, chosen per policy and overridable per entity type.
Policies you can bind anywhere
Versioned policies bind to users, teams, roles, organizations, AI sources and pools. Choose to redact and continue, or block the message with an edit-and-resend path.
Audited token vault
Protected values stay masked until a permission-gated, audited reveal, with click-to-reveal chips in chat for people who are allowed to see them.
Custom recognizers and profiles
Add organization-specific detection patterns with a live test bench, or name a packaged HIPAA Safe Harbor, GDPR or PCI profile in a policy.
How it works
- 1
Build a protection policy
Choose a confidence threshold, entities and language, then an operator per entity: replace, redact, mask, hash, synthetic fake, tokenize or encrypt. Name a HIPAA, GDPR or PCI profile if needed.
- 2
Bind it to who and what
Attach the policy to users, teams, roles, organizations, AI sources, pools or Warden keys. Decide whether detection redacts and continues or blocks the message with an edit-and-resend path.
- 3
Protect traffic in flight
Personal data is redacted before the request reaches a model provider. Senders see per-message indicators, and the response shows originals restored, kept redacted, synthetic or revealed on demand.
- 4
Review detections and reveals
Reversible values sit in a token vault, shown only through an audited, permission-gated reveal. The dashboard reports detections, blocked messages and reveals; anyone can file a false-positive report.
Ask AI about Shroud
Every Shroud page has an Ask AI button. It opens a chat that already has the page, its rows, and your selection. Try asking:
- Which protection policies here block messages instead of redacting and continuing?
- Which entity types appear most often in this vault list?
- Which recognizers are custom, and what entity type does each detect?
How teams use it
- Stop customer personal data from reaching an external model provider.
- Scan uploaded documents for personal data before they are stored.
- Let compliance staff reveal a tokenized value with a permission check and an audit record.
Frequently asked questions
See Shroud in action
Join the early-access list, or talk to our team about your environment.
Talk to us