01
Layer 01
Privacy Filter
Personal data (names, addresses, IBAN), bank details, medical diagnoses — masked or blocked before reaching the AI model.
PIIIBANMedical
MultiLayer Data Shield · 5 Layers
Five layers between your team and the AI.
Every employee request runs through configurable inspection layers — before reaching an AI model. Sensitive data detected and masked, problematic content filtered. Effectiveness and responses are controlled via your admin configuration.
PII maskingJailbreak protectionAudit trail
p95 ≈ 442msSafety-Layer
9 / 9Risk categories
Made in Austria
Made for Europe
Protection at every layerGDPR Art. 28EU AI Act 2026Encrypted at-rest + in-transitAudit trail of all requestsEU-only Hosting
MultiLayer Data Shield
The five protection layers.
Each layer operates independently — multi-layer inspection reduces the risk that individual content slips through unnoticed.
02
Layer 02
Content Filter
Insults, discrimination, illegal content, drug references — requests are blocked or sent for confirmation depending on configuration.
HateDrugsExtremism
03
Layer 03
Manipulation & Behavior Filter
Self-harm, prompt-injection attempts ("Ignore all instructions…"), harmful requests — blocked or sent for confirmation, depending on configuration.
ManipulationEthicsJailbreak
04
Layer 04
Audit & Logging
Every interaction is logged — as a basis for your own compliance evidence, audits and reports.
AuditGDPRReports
05
Layer 05 · NEW
Anti-Hallucination Layer (Veritas Chat)
On critical questions the answer is checked by 5 independent AI models, a leader filters the consensus. Single hallucinations don't pass — disagreements are shown transparently.
Consensus5 AIsAnti-Hallucination
What we detect
Detection categories.
Names & Personal Data
Email Addresses
Phone Numbers
Credit Card Numbers
IBAN / Bank Details
Social Security Numbers
Addresses
Medical Data
Comparison
With vs. without protection.
Status quo
Without protection
- × Sensitive data flows uncontrolled to AI models
- × No audit trail for compliance
- × Data loss goes unnoticed
- × Shadow AI via private accounts
HOVIGuard
With MultiLayer Data Shield
- ✓ PII detected and masked in real time
- ✓ Complete audit trail for every interaction
- ✓ Admins maintain full control
- ✓ Central platform instead of Shadow AI
- ✓ Veritas Chat: 5 AIs check answers — anti-hallucination
FAQ
Frequently asked questions.
Can I customize detection?
Yes. Admins can configure PII categories and strictness levels per user group.
What happens to detected data?
Detected PII is masked or blocked depending on configuration. Detected values are replaced with placeholders before the AI call; the masked request is sent to the model.
Does it work with uploads too?
Yes. Documents are analyzed before processing. Admins control allowed file types per group.
How does the 5th layer (Veritas) differ from the other four?
Layers 1-4 check the INPUT before it's sent to the AI (privacy, content, manipulation, audit). Veritas (layer 5) checks the OUTPUT — by querying 5 independent models in parallel and synthesising consensus. Pre-model and post-model protection interlock.
Is Veritas Chat active in every plan?
Pro and Enterprise: yes, at no extra cost (uses the normal fair-use budget). Demo: yes, with limited budget. Company admins can disable Veritas per tenant or per user. Details on the Veritas Chat feature page.