AI privacy

What this page is about

When you work with our tools, an AI model sometimes runs in the background. These models run at specialized providers and only see what we send them. Before text goes there, we replace what our system recognizes as personal. A few paths skip this step; they are listed below with the providers.

What happens to your data before it reaches the AI

Imagine you ask our software: “Write an email to Hans Müller, hans@beispiel.de.” Before this request goes to the AI, our system replaces the name and the email address with invented stand-ins. The AI sees something like: “Write an email to Jamie Kozakiewicz, maxwell_huels@jacobi-luettgen.example.” The model does not know it was about Hans Müller. When the answer comes back, we swap the stand-ins back. You see the real details, the AI only the stand-ins. That is how the chat works. The knowledge search and the tools for other programs replace with placeholders like [PERSON:…] instead.

This is called pseudonymization. Here is what we replace before anything goes to the AI.

People
First and last names, email addresses, phone numbers
Addresses
Postal addresses, dates of birth, places of birth
Finance
IBAN, BIC, credit card numbers
IDs
ID card and passport numbers, tax IDs, social security numbers

The link between the real value and the stand-in stays in our system only. The AI never gets to see it.

Which AI providers we use

Every request to an AI model goes through Cortecs, with two exceptions at Google: web search in the chat and the image description (alt text) for blog images. Google text-to-speech also reads blog posts aloud. When the chat fetches a web page (page fetch), its address goes to that website in plain text. Every chat answer that used web search or page fetch carries a notice saying so. Your organization can switch web search and page fetch off in its settings under Privacy. Every single call leaves an entry in our audit log. Which data processing agreement applies to which provider, how long it stores data and where it processes it is listed with sources on the page AI transparency.

Cortecs

Routing to AI models (Vienna)

Google

Web search in the chat and alt texts for blog images (Gemini) plus reading blog posts aloud, each without pseudonymization

What we can prove to you

Trust is good, evidence is better. That is why we built an audit log.

Every AI call has an entry

Who asked when, which provider answered, how many tokens were used. The question itself is not in it.

The log is protected against tampering

A Postgres trigger blocks every UPDATE on the audit table. DELETE is allowed only for a narrowly scoped cleanup role that exists solely to enforce the retention periods. So we cannot change recorded entries afterwards or remove them one by one.

Daily integrity check

Every day a hash chain check runs over all entries. The result is in the daily email to the administrators.

Right to be forgotten, with proof (Art. 17 GDPR)

If you agree to or request the deletion of your AI-related data, we delete the mappings atomically and record the process in several steps: a start entry, one step entry per source, a completion entry. We keep these deletion records for three years. That way we can still prove years later that your data is gone.

What we cannot promise

Honesty is part of it. Technically, we cannot rule out these three points completely.

AI providers see stand-ins. We trust them by contract not to trace these back to personal data. We cannot prevent that technically.

Our recognition works well, but it is not perfect. Rare names or foreign formats in particular can slip through. We test this systematically and keep improving it.

When a member of our support team helps you, they see your data in plain text. That is intended and necessary, otherwise nobody could help you.

If you have questions

Do you want to know what the AI actually saw about you? Would you like your data deleted and a record of it? Write to us. We answer as people, not as a form.

Data protection officer, Michael SCHILLER - Organisation. Digital.

datenschutz@schiller-partners.de

Last updated: 2 October 2026