AI in mid-sized companies: work productively without company data leaking out

What is the actual risk — and what is not?

The risk is not "the AI". It is the body of data that builds up at a provider that stores: quotes, calculations, customer data, source code, strategy papers — once copied into a public chatbot, they are outside your control. The best-known example: in 2023 it emerged that Samsung employees had entered internal source code into ChatGPT — the company subsequently blocked its use. On top of that comes the legal position: personal data (customers, employees) requires a data processing agreement under the GDPR and, with US providers, an answer to the CLOUD Act and Schrems II. And from here the clock of the EU AI Act is running, which has applied in stages since 2024 and brings transparency and training obligations among others.

The second, underestimated risk is called shadow AI: if the company provides no official tool, employees use private chatbot accounts — uncontrolled and invisible. The solution is not prohibition but a compliant channel.

What do mid-sized companies actually use AI for?

1. Sales & quotes: quote texts, tender responses, follow-up emails — ready-to-send drafts from bullet points.
2. Correspondence: customer enquiries, complaint responses, supplier communication — faster and more consistent, in multiple languages too.
3. Internal documents: minutes, work instructions, training materials, quality management documentation.
4. Summarising & understanding: condensing long contracts, standards, tender documents and market reports to the essentials.
5. Development & IT: code assistance via the API — without source code becoming training material for someone else's model.

Industries with particular confidentiality obligations have their own pages: law firms, tax advisors, healthcare, HR departments.

How PrivatAI works

PrivatAI stores nothing. Prompts and responses are processed and discarded — no content logs, no training, no body of data. One exception: in the event of system errors or detected misuse, the inference data centre may temporarily retain a single request in order to analyse the cause; this is contractually limited and disclosed in the DPA (§ 4 no. 2). The application runs in Germany (Hetzner), the AI processing in France (Scaleway): your data stays in the EU and US access rights do not apply. Two ways in:

• Browser chat — the official, compliant replacement for private chatbot accounts. Work is done by pasting text in; there is no file upload yet. Everyone who needs it gets their own account: there is currently no central user administration and no shared team workspace, so a rollout means one subscription per person.
• OpenAI-compatible API for your systems: any application that speaks the OpenAI interface — from an internal tool to an AI agent — can be switched to PrivatAI by swapping the endpoint and API key. An embeddings endpoint is included, so retrieval over your own document store can be built on top of it.

DPA under Art. 28 GDPR including documented technical and organisational measures (TOMs); for professionals bound by secrecy, with a § 203 clause and a separate confidentiality undertaking that binds PrivatAI itself. Passing that obligation down the entire sub-processor chain, and AI inference in Germany, are not part of Essential or Professional but of the separately agreed Enterprise setup.

PrivatAI (privatai.com) — AI chat & API from Germany. No content logs*, no training on your data, GDPR-compliant.

PrivatAI compared (default settings)

All figures describe normal operation. The one narrow exception — temporary retention of individual requests on system errors or detected misuse — is disclosed in full in the DPA (§ 4 no. 2).