Articles on AI in the enterprise. Each one first answers a question in a few sentences, then explains what it looks like in practice.
Published
Is it enough to tell an AI assistant in its prompt not to reveal confidential content?
No. An instruction in the prompt is a request to the language model, not a rule: what the model has read, it can repeat. Only a fixed, verifiable rule outside the model is reliable, one that decides before the search which documents a request can reach at all.
Published
Can a company use AI without its data leaving the building or the EU?
Yes. A knowledge search or an assistant can be built so that data, search and language model run in your own environment: in your own cloud account in an EU region, in your own data centre or on your own hardware. The question is not whether AI is used, but where it runs and who decides what it sees.
Published
How does software stay secure and current after the handover without unreviewed changes going live?
By being built from the start to measure continuously whether it does what it should, to detect deviations and newly known security flaws itself, and to prepare and test corrections and updates automatically. A change only goes live once you have approved it, with traceable evidence.
Published
What should an AI knowledge search do when its documents hold no answer?
Decline. A reliable knowledge search backs every answer with its source, that is file, page and line, and says openly when the corpus holds no passage that supports the answer. A fluent answer without evidence is more dangerous than none, because nobody notices that it is wrong.