The pressure to adopt AI is real. Boards want productivity gains, staff are already experimenting with ChatGPT, and tools such as Microsoft 365 Copilot can be switched on in an afternoon. But there is a question that too few organisations ask first: what data will the AI be able to see?
AI assistants work by searching, summarising and reasoning over the information you already hold — your file shares, SharePoint sites, Teams chats, mailboxes and line-of-business systems. If that information is ungoverned, the AI inherits every problem in it. It does not create your data problems. It finds them, and then presents them back to your people with confidence.
AI doesn’t create your data problems — it surfaces them
Most enterprise AI tools respect the permissions you already have in place. That sounds reassuring until you look at what those permissions really are: the HR folder shared with “Everyone” years ago, the finance workbook sent as an open link, the project site nobody tidied up when the client left, the leaver whose access was never removed. Before AI, these were risks hidden by obscurity — finding the file meant knowing where to look. With an AI assistant, anyone can simply ask.
The result is oversharing at the speed of a prompt: salary details, disciplinary records, client contracts or personal information appearing in answers to people who should never have seen them.
Five questions to answer before you switch AI on
- What data do we have, and where is it? Across Microsoft 365, file servers, cloud applications, personal OneDrives and the shadow IT that grew up around them.
- Who can access it — and who actually uses it? Permissions and real usage are rarely the same thing. Data that nobody has opened in years but dozens of people can reach is pure risk.
- How old is it, and is it still accurate? Outdated policies, superseded price lists and old drafts sit alongside the current versions. AI cannot reliably tell which one is authoritative.
- Why are we keeping it? Every set of personal data needs a legitimate purpose and a lawful basis — and a point at which it should be deleted.
- How sensitive is it? Personal data, special category data such as health information, commercially sensitive material and client-confidential information all need different handling.
Problem one: mixed results
AI is only as good as the information it draws on. Feed it redundant, obsolete and trivial data — the duplicates, the old versions, the “final-v3-FINAL” documents — and you get answers that are confidently wrong or inconsistent from one day to the next. Staff lose trust quickly, adoption stalls, and the business case that justified the investment never materialises. The problem gets blamed on the AI, when the real cause is the data underneath it.
Problem two: you may be breaching UK GDPR
The bigger risk is regulatory. UK GDPR requires personal data to be collected for specified purposes, limited to what is necessary, kept accurate, kept no longer than needed and kept secure — and it requires you to be able to demonstrate all of that. Pointing AI at data you don’t understand can put you on the wrong side of several of those principles at once:
- Purpose limitation: using personal data collected for one reason in a new AI process may need a fresh assessment of whether that use is compatible and lawful.
- Storage limitation: data you should have deleted years ago is now being actively processed and surfaced.
- Security and confidentiality: overshared personal data becomes visible to people with no reason to see it.
- Accountability: if you don’t know what you hold, you can’t answer a subject access request fully or show a regulator how you manage risk.
New uses of personal data with AI will often call for a Data Protection Impact Assessment, and the Information Commissioner’s Office has been clear that organisations remain responsible for how AI uses the personal data they hold. Serious breaches can lead to fines of up to £17.5 million or 4% of annual worldwide turnover, and the reputational damage of an AI tool exposing personal information is harder still to repair.
Go in with your eyes open
None of this means waiting. It means sequencing the work so AI lands on a foundation you understand:
- Discover what data you hold across the organisation and where it lives.
- Understand who has access, who uses it, how old it is and how sensitive it is.
- Decide what you have a legitimate reason to keep, and set retention rules for it.
- Remediate — remove what you no longer need, archive what you must keep, and fix oversharing.
- Pilot AI on a governed scope, measure the results and expand with confidence.
Done this way, AI readiness becomes a business benefit in its own right: lower storage costs, a smaller attack surface, faster subject access responses and better answers from the AI when you switch it on.
How Rockmore can help
Rockmore’s Data Discovery Service gives you an independent, evidence-based view of your data before AI goes anywhere near it. We map what you hold and where, review access and oversharing, analyse the age and use of your data, identify personal and sensitive information, and test what you keep against the legitimate reasons for keeping it. You receive a clear, prioritised roadmap: what to fix first, what to delete, and where AI can be introduced safely.
We are vendor-neutral, so our advice isn’t shaped by a product to sell — and we work alongside your team or your MSP to put the recommendations into practice.
If AI is on your roadmap, start with your data. Talk to Rockmore about our Data Discovery Service.