AI and data processing
Where data is processed, what we promise, and when we advise against AI.
Draft. This text needs review by a lawyer before launch.
We advise against AI more often than we recommend it
If a result has to be explainable, auditable and identical every time, a language model is the wrong tool. Then we build arithmetic instead. We say so before a quote is written.
Where data is processed
We use Anthropic's Claude API. Data is sent for processing and is not used to train models. For client work we agree specifically which data may be sent and what stays with you — and we sign a data processing agreement before touching personal data.
Recording and transcribing calls
Recording a conversation you take part in is lawful in Denmark. But processing the recording afterwards is processing of personal data, which needs a lawful basis and requires informing the participants. We help get that in place — but we are not lawyers, and we recommend your solicitor reviews your setup.
Transparency towards your customers
If we build something your customers talk to, it must be clear that it is an AI system. That follows from Article 50 of the EU AI Act, and we build it in from the start rather than bolting it on afterwards.
The demos on this site
What you type into the demos is not stored and not logged. They are protected against abuse with a cap on how often they can be used, but they are not covered by a data processing agreement. So don't paste personal data or confidential material into them.
Last updated: 2026-08-13