Security
Security and privacy: where your data sits and who sees it
For every project, UNID AI states in writing where data is stored, which data goes to which language model and who has access. Predict can run fully on your infrastructure. With Second Brain, archive, index and records stay on the server; only the passages an answer needs go to the AI assistant.
Last updated:
Data flow per product and deployment
| Product · deployment | Stored | Sent to a language model | Processors |
|---|---|---|---|
| Second Brain · controlled server | Documents, search index, records, log on the server | Only the passages for each answer, to the agreed AI assistant | Server operator and model provider, as agreed |
| Predict · your infrastructure | Everything in your data centre | Nothing outside; local language model | No external ones |
| Predict · with cloud model | Data stays in your database | Question and query results (e.g. sums, rankings) | Model provider with Art. 28 GDPR contract |
Which model provider, which region and which contract terms apply is agreed in writing for each project.
Access and logging
- Every person signs in; no shared accounts
- Permissions per area: everyone sees only what they are authorised for
- Log of every access and every query
- Predict with read-only access and row limits
- Drafts and decisions need human sign-off
GDPR
- Processing agreement under Art. 28 GDPR when we process data for you
- Data minimisation: predictions need no names or addresses; pseudonymisation is standard
- No special categories of personal data (e.g. health data) in our standard use cases
- Deletion or return of all data at the end of the contract
EU AI Act
The EU AI Act (Regulation (EU) 2024/1689) distinguishes providers and deployers of AI systems. As the company using the system you are usually the deployer; we support you with documentation on how it works, the data and its limits.
Whether a use counts as high-risk depends on its purpose. Annex III lists, for example, risk assessment and pricing in life and health insurance. A knowledge assistant for documents or a retention ranking is usually assessed differently, but needs your own assessment.
Dates: the Regulation has been in force since 1 August 2024; prohibitions and the AI literacy duty (Art. 4) apply since 2 February 2025. Obligations for Annex III high-risk systems were postponed in 2026 by the so-called Digital Omnibus package, reportedly to 2 December 2027. The text in force in the Official Journal is authoritative.
EU AI Act on EUR-Lex ↗ · RTR AI Service Desk (Austria) ↗
General information, not legal advice. As of 22 November 2026.
Works council and human oversight
Systems that change how people work are discussed early with the works council and the data protection officer. The AI suggests; people decide. Every use case is designed that way.
What we do not claim
- We currently hold no certification (e.g. ISO 27001) and claim none.
- We do not call anything “GDPR-certified”; no such general certification exists.
- We do not write “no data leaves the building” when a cloud model is involved.
FAQ
Frequently asked questions
What does GDPR-compliant AI mean in practice?
What does on-premise mean for AI?
Is our data used to train models?
Which EU AI Act deadlines apply to us?
Who is a processor?
What happens to the data at the end of the contract?
Questions from IT or data protection?
We are happy to answer them directly with your IT security team or data protection officer, including a data-flow diagram for your project.
UNID AI · Leonard-Bernstein-Straße 10 · 1220 Vienna · +43 660 476 30 89