We Use Cookies

    We use cookies to enhance your browsing experience, analyze site traffic, and personalize content. By clicking "Accept All", you consent to our use of cookies. You can customize your preferences or reject non-essential cookies.

    Learn more about our cookie policy
    Privacy & Security

    PII Guard Update: One Toggle, Fully Sealed

    PII Guard is now a single account-wide setting — and with it, we closed the last gap: PDF attachments in chat.

    Basics
    4 min read
    Update

    What Changed

    PII Guard used to be a per-model toggle. In practice, protection is an account-level decision — either your data is sensitive or it isn't, regardless of which model you talk to. So we simplified it: one PII Guard toggle in your Account Settings (Model Library) now controls protection for ALL models. And we went one step further to fully seal the space: while PII Guard is on, PDF attachments in chat are disabled, because they would be processed by an external OCR service before any redaction could happen.

    One Toggle for All Models

    Enable PII Guard once and it applies to every model in your library — including models you add later. No more per-model bookkeeping, no risk of forgetting a toggle on a newly added model. You find the toggle at the top of your Model Library in Account Settings. The +30% credit uplift for Presidio processing applies per protected request, unchanged.

    Chat Stays Fully Protected

    Everything you type or paste into the chat is still redacted before it reaches any AI model: Presidio scans your text for names, email addresses, phone numbers, IBANs, medical identifiers and more, and replaces them with neutral placeholders. Copy & paste of sensitive text is safe. The model answers your question — it just never sees the raw data.

    Chat PDFs: Blocked by Design

    A PDF attached directly in a chat cannot be redacted before OCR — the document would have to be read by an external OCR engine first, and at that point unredacted content would have left the protected pipeline. Rather than accept that gap, PII Guard now blocks PDF attachments in chat entirely. You'll see a clear message pointing you to the safe path: the Knowledge Base. Nothing leaves — a block cannot half-fail. If PII Guard is OFF, chat PDF uploads work exactly as before (EU-hosted Mistral OCR).

    The Safe Path: PII-Protected Knowledge Base

    For sensitive documents, upload the PDF to a Knowledge Base with the PII filter enabled and reference it in chat with "#". The document is processed entirely on-premise: Marker (self-hosted GPU OCR) extracts the text, Presidio redacts all PII, and only the clean text enters your Knowledge Base. No raw document content ever reaches an external service — and you keep full control over which documents are referenced in which chat.

    Read the full PII redaction deep-dive →

    Behaviour at a Glance

    What happens with the toggle on and off.

    PathPII Guard OFFPII Guard ON
    Chat text (typed / pasted)Sent to the model as-isRedacted by Presidio before reaching any model (+30% credits)
    PDF attached in chatProcessed normally (EU-hosted Mistral OCR)Blocked — use a PII-enabled Knowledge Base instead
    PDF uploaded to Knowledge Base (PII filter on)On-premise Marker OCR + Presidio redactionOn-premise Marker OCR + Presidio redaction (unchanged)

    Why This Is the Right Trade-Off

    A redaction pipeline can fail halfway — a block cannot. By rejecting chat PDFs while PII Guard is active, we guarantee that no unredacted document ever leaves the protected infrastructure. The Knowledge Base path gives you the same capability with a stronger guarantee: fully on-premise OCR and redaction, covered by our Zero Data Retention Policy and GDPR framework. As always: Presidio runs on our European servers, redacted values are never stored, and PII-protected usage is transparently labelled in your dashboard.

    Frequently Asked Questions

    PII redaction powered by Microsoft Presidio, running on AI·Collab's European infrastructure. Covered by our Zero Data Retention Policy (ZDRP).

    Related Articles

    Ready to Experience 300+ AI Models?

    Get started today. Access models from OpenAI, Google, Anthropic, Grok and more.

    GDPR compliant · Zero data retention · Cancel anytime