Protect sensitive inputs before you send them.

Local protection

Keep the meaning. Replace the sensitive details.

Select the inputs and documents you want to protect. Sano Shield detects sensitive values and replaces them with placeholders on your device. The AI model receives those placeholders with the surrounding context.

When the response comes back, Sano can restore the protected values locally. Your mappings remain on your computer throughout.

• Choose what to protect
• Detect and replace on your device
• Preserve context for the task
• Restore values in the answer

Sano Shield 1

Measured in German and English.

Sano Shield 1.2 is the detection model behind the protection layer. On paired held-out synthetic stress suites of 3,000 records per language, run through Sano’s current product pipeline, required-span recall is 98.30% in German and 96.32% in English. Combined required-span recall is 97.29%. The evaluation data was never used for training.

Explore the model card for the evaluation method, supported categories and release details. Review detected values before using prepared content for sensitive work.

• 98.30% German required-span recall in the stated suites
• 96.32% English required-span recall in the stated suites
• Published model card and evaluation
• Detection runs locally

Hosting & privacy

Your work stays your business.

Choose German or EU-hosted models in Sano. Every model comes with zero data retention and no training on your prompts or outputs. German-hosted models also come with contractual confidentiality under §203.

Your conversations and privacy mappings stay on your device. Sano Shield detects and replaces sensitive values locally in the inputs you select, then restores them in the response.

• German or EU-hosted models
• ZDR and no model training across all models
• §203 confidentiality for German-hosted models
• Local conversations and control over agent actions

Select. Protect. Create.

01 · Select your inputs

Choose the text or documents you want to protect.

02 · Protect locally

Review the detected values and replace them before the model request.

03 · Use the response

Restore protected values locally and continue working with the result.

AI data protection questions

How does local input protection work?

Sano Shield protects the inputs you select. Agents then work with your prepared files. Tool results, code, browser content and excerpts from connected services can reach the selected model without automatic anonymization.

Does Sano Shield detect every sensitive value?

Detection has a measured scope and can miss values. Review the prepared input; the published model card explains the evaluated categories and results.

Where can I read the evaluation?

The SanoAI model card on Hugging Face documents Sano Shield 1.2 and its German and English evaluation. Follow the model-card link on the homepage.

Where are my AI requests processed?

Choose German or EU-hosted models in Sano. Every model comes with zero data retention and no training on your prompts or outputs. German-hosted models also come with contractual confidentiality under §203.

More ways to work with Sano

Privacy-First AI Workspace

See how local protection fits into everyday AI work.

Secure AI for Enterprise

Plan protected AI adoption across teams and organizational controls.

Put local protection to work.

Tell us which documents and sensitive inputs you work with. We’ll walk you through Sano Shield in Sano Work.