Today, there’s no denying it: AI has moved to the very heart of how companies operate. Behind every prompt, there’s internal data. So how do we deal with sovereignty?
AI stopped being a mere trend a long time ago, it’s part of daily life now.
The thing is, every time we use AI, we’re handing a model information that travels through servers subject to foreign jurisdictions (American ones), in most cases.
Some people are fully aware of this reality, others play it down, given everything these tools bring to the table.
So it’s fair to ask: how do we deal with it?
There’s no ready-made answer, but there are habits worth adopting to protect yourself a little better.
- Anonymise your data as much as possible before submitting it to a model. Never include proper names, client names, or identifiable internal references. Work with variables, letters and codes instead.
- Decide case by case, not everything warrants the same level of caution. Drafting a follow-up email with a consumer-grade tool? Limited risk. Analysing an acquisition strategy using client data? I’ll let you be the judge.
- Ask yourself which AI, for what, and with which data. There are now industry-specific models, enterprise AI deployed in-house, solutions hosted in Europe… Tools like Mistral AI or LightOn don’t always match the large American models on performance, but they significantly reduce your exposure.
Behind all these recommendations, the real point is recognising that AI is both a market and a geopolitical lever.
When an AI provider has to cut off access overnight to comply with a foreign government’s export controls (as was the case with Anthropic’s Fable 5), you realise that technological dependence has very real consequences.
Digital sovereignty isn’t just about data centres and undersea cables. It also lives in these small, everyday decisions.
This news was first published on LinkedIn. To comment or share it, head to the original post. → See the post