This website works with AI: it answers questions, writes sections on and collects what you looked at for the contact form. That only works if it is clear where which data ends up. Here it is, without legalese.
In short
- Hosting in Frankfurt: website, server functions and storage run on Amazon Web Services in Frankfurt am Main.
- Encrypted throughout: HTTPS only from TLS 1.2, all storage encrypted and not publicly reachable.
- No cookies, no third-party tracking: no cookie banner, no third-party analytics, no recognition on your next visit. We only count anonymously which topics are asked about.
- AI models only in EU data centres: every request to a language model is processed in a data centre in the EU. The connection runs through the Frankfurt gateway of our partner Requesty; the contractual partner is Requesty Ltd, based in England - the EU has an adequacy decision for the United Kingdom.
- Nothing stays with the model provider: zero data retention - your inputs are not stored there, not used for training and not published or shared.
- Your visit stays in your browser tab: what the website remembers for you is kept in your browser’s session storage and is gone when you close the tab. Exceptions: three settings you change yourself (“show instantly”, “No more tips”, “Think along”) - they stay in your browser until you reset them. Questions to the AI assistant are stored for 30 days, cleaned of recognisable contact details, plus technical logs without content (see below).
Where the website runs
The pages themselves are pre-built files. They are stored on Amazon Web Services in Frankfurt and delivered through the Amazon CloudFront content delivery network so that they load fast. For us, CloudFront only uses locations in Europe, North America and Israel; visitors from Europe are usually served from European locations. We do not analyse access logs.
Everything that computes - the AI assistant, writing on, the contact form - also runs as a server function in Frankfurt. Fonts are loaded from our own server too; there is no connection to Google or other font providers.
Which AI models we use
We only use language models in data centres in the EU. For this website, our only contractual partner is Requesty; Requesty engages the model providers behind it under its own responsibility:
- For answers and texts: language models in data centres in the EU.
- For selection and classification, such as which content fits next, and for checking every assistant answer before it is shown: a decision model, also in the EU. It writes no text and only returns probabilities. If it cannot be reached, the answer appears unchecked and is recorded as such.
- The connection runs through our partner Requesty (Requesty Ltd, based in England - the EU has an adequacy decision for the United Kingdom) and its EU gateway in Frankfurt. Our organisation is set up there with fixed defaults: only this EU gateway, zero data retention, access keys without logging and only models that meet all the conditions below - the website cannot call any others. Requesty itself stores no content: logging is off and zero data retention applies there too. This is a deliberate choice: one central gateway where our rules are technically enforced for the whole organisation - EU gateway, zero data retention, allow list - instead of many separate contracts with different terms. Through Requesty we use Microsoft Azure and Amazon Bedrock, among others, as well as European providers such as Mistral AI. We are happy to share the full list on request.
For all models: zero data retention - the provider does not store your inputs and the answers beyond processing. No training - your inputs are not used to improve models. No sharing - inputs are neither published nor passed on to third parties.
Only what is needed for the one answer goes to the models: the title of the page, your path to it and what you confirmed - or, in the assistant, your question.
What we store ourselves
Zero data retention applies at the model provider. We store little ourselves, and only as long as needed:
| What | Where | How long |
|---|---|---|
| Assistant conversations, cleaned of recognisable contact details - email addresses and phone numbers are filtered out automatically first, names in free text may remain; not linked to the IP address. Plus the result of the check before display | Frankfurt, encrypted | 30 days, for quality assurance |
| Technical logs without conversation or form content: error code, reference number, duration. On incidents the system alerts our operations team | Frankfurt | 30 days |
| Counters for the limits of the AI features: hashed session number or hashed IP address, minute, count (more) | Frankfurt | 2 minutes |
| Anonymous counts of topics and path patterns, without wording | Frankfurt | as statistics |
| Your request via the contact form | as an email to our team, sent from Frankfurt | until handled, then according to our deletion rules |
| Magazine editions of our own articles and references | Frankfurt | until the next change - public content only |
Encrypted in transit
Every connection is encrypted - from the browser to the language model:
- HTTPS only: requests via unencrypted HTTP are redirected to HTTPS. Only current protocols from TLS 1.2 upwards are allowed.
- HSTS: your browser remembers for one year to open this website and its subdomains only over an encrypted connection - even if someone shares an HTTP link.
- Encrypted internally too: between the delivery network and the server functions, to the storage and to the language models, everything also runs over TLS. The storage rejects unencrypted access as a rule.
- Protection in the browser: security headers prevent the website from being embedded in other sites and only pass on the domain, not the full address, when you leave.
Encrypted and locked down at rest
- Encrypted at rest: all storage at AWS is encrypted - conversations and magazine editions with keys from the AWS Key Management Service, the statistics tables and website files with AWS’s own encryption.
- Not publicly reachable: no storage is public. Only the delivery network reads the website files, via signed requests; the server functions only accept requests that demonstrably come through this network.
- Least privilege: each server function may only read and write exactly what it needs. Credentials, for example for the language models, are stored encrypted in the AWS Parameter Store, not in the code.
- Deleted automatically: AWS deletes conversations and logs by itself after 30 days - nobody has to remember to do it.
What stays in your browser
So that the website can think along, from your first interaction it remembers in your browser’s session storage which pages you read, what you ask and what you answer. This applies to this tab only, can be seen in full under Your focus, deleted any time with “Forget everything” and disappears when you close the tab - except for the three settings above that you change yourself. With “Think along” off, it does not remember your path at all. The blocks in the contact form come from there too: you can take any of them out, and the selection only reaches us if you send it.
Being straight about the CLOUD Act
The website runs on Amazon Web Services, a US company, and model providers behind Requesty can be US companies too. Even though all data centres are in the EU, US authorities can in principle request data from US providers under the CLOUD Act. We say this openly: it is the status quo for almost all European companies using cloud and AI today. We are working on moving more of our infrastructure to providers based exclusively in the EU. In projects, models from European providers can already be used through Requesty today. Until then we keep the amount of data that reaches them as small as possible - see above.
And in your project?
This page describes our website. In client projects we follow your setup:
- Your environment or ours: depending on the project, we work in your infrastructure - your cloud, your models, on-premises up to environments without internet access (air gap) - or in our own environment. Local models trail the cloud by a few months, but are already good enough for many well-defined tasks (the gap in numbers). Where classified information is involved, security-cleared staff work on the project. For classified or export-controlled data, only your environment is used. If you already have your own cloud tenant, for example on Microsoft Azure, we are happy to use it.
- Models and tools: which models and tools are used and where they run, we decide with you project by project. The rules above apply to this website - we recommend the same for your project and use them as the default unless you ask for something else.
- Information security: SII Germany holds the TISAX label at the highest level AL3, including prototype protection and the secured premises it requires.
- Contracts: by default we sign a data processing agreement under Art. 28 GDPR, and a non-disclosure agreement on request. Which sub-processors are involved depends on the project and is agreed with you; the only fixed one is Microsoft 365 for our own collaboration, except for classified data.
- Teams outside the EU: we only involve them with your consent, on the basis of the EU standard contractual clauses.
- Security incidents: we report incidents affecting your data to your designated contact within 24 hours.
- Options we know and use: see the overview below - options, not defaults.
| Tool | Used for | Your data | Contract with |
|---|---|---|---|
| Language models through Requesty | Development and solutions | EU data centres, zero data retention, only models on our allow list | Requesty - our recommendation |
| GitHub Copilot | Development in projects | Business licence, no training on your code | GitHub (Microsoft) |
| Microsoft Copilot | Teamwork | within the company's Microsoft 365 tenant, no training on your data | Microsoft |
Requesty bundles a wide range of model providers with data centres in the EU, including European providers; our contractual partner is Requesty, not the individual model providers. We use GitHub Copilot because of its wide adoption and acceptance, backed by Microsoft. We still recommend Requesty, because it goes further in ensuring EU data sovereignty. We work with Microsoft, OpenAI and Anthropic as technology partners.
What applies to your project we clarify in the first conversation.
What we deliberately do not do
- no cookies and no cookie banner
- no third-party analytics, tracking or marketing tools - only our own anonymous count without an identifier
- no identifier, no account, no recognition on your next visit
- no profile of you beyond the current tab
How the website thinks along with this data is described in How this website thinks along. The legal details are in the privacy policy.
Found a vulnerability?
If you discover a vulnerability on this website, please write to informationssicherheit@sii-germany.com. We confirm receipt, review the report and get back to you with the result. The address is also available in machine-readable form in the security.txt file.
