AI on top of technical debt
Oct 29th, 2025 by Jesper Kråkhede
Everyone wants to introduce AI. Few talk about what it is supposed to run on.
AI assistants and agents work with the data and permissions they get. In many organisations, that data lives in file shares with access rights nobody has reviewed in ten years, in SharePoint sites shared with ‘everyone’, in old systems without modern authentication and in mailboxes full of documents that should have been deleted long ago.
Before AI, much of that was hidden by obscurity. Nobody found the confidential document in the old project folder because nobody looked. An AI assistant looks. It will happily summarise the salary list it found in a folder that was accidentally shared with the whole company. Overshared data becomes a lot more visible.
The same goes for identity and access. If your admins have too many rights, so do the agents you build on top of them. If your service accounts are a mess, adding agents will make it worse.
So for management: Treat AI as a reason to fix your foundations, not as a way around them. Before rolling out AI broadly, clean up permissions on the most sensitive data. Label it, so the AI knows what it may and may not use. Review who has access to what. Fix the identity basics.
It isn´t glamorous. It is what makes AI safe.
AI doesn´t create your data problems. It makes them impossible to ignore.