Skip to content

AI & Data Governance

Make your data safe for AI.

An AI assistant can draw on everything a user can open. We find the files shared too widely, fix access and labels, and set the rules, so your AI rollout can go ahead safely.

Built by the team behind the data platforms of

The problem

Your AI finds what was only hidden.

  • Hidden, not protected

    Files shared with everyone years ago were safe only because nobody knew where to look. An AI assistant does know: it searches everything a user can open.

  • The rollout stops

    One employee finds the salary sheet through Copilot, and the pilot is paused while security, the works council and the privacy officer ask questions.

  • Nobody can say what is sensitive

    Sites have no owner, files have no label and access was granted per person. Nobody can say what the AI may read, so nobody dares to switch it on.

Copilot

Rollout paused
Sales employee

What do people in my team earn?

Here is an overview of the salaries in your team, based on:

Salaries 2019 final.xlsxOld project siteShared with: Everyone

This file was never meant for you

1,240files shared with everyone, found in one scan

Oversharing scan

See what your AI can see.

Before you widen an AI rollout, we scan SharePoint, Teams and OneDrive for what any user, and so any AI assistant, can open. You get every finding with its risk and its fix. Pick a finding to see an example.

Example: a company of 450 people, halfway through its Copilot pilot

Items to fix before the next wave5,593

What we found

Sites and folders open to every employee, set up years ago to make working together easy. Among them: salary sheets, contracts and a performance review folder.

Why it matters for AI

Every employee can open them, so every AI assistant can quote them in an answer.

How we fix it

We replace “everyone” with the team’s own group, and the site owner confirms who really needs access.

Clean-up

Access fixed at the source.

We do not hide the problem behind a filter. We fix who can open each file and label what is sensitive, so every AI assistant, search and colleague sees the same, correct picture. Switch between before and after.

Files the AI would read
  • Salaries 2019 final.xlsxOld project siteEveryoneNo label
  • Absence records 2024.xlsxHR archiveEveryoneNo label
  • Board minutes March.docxOld board siteEveryoneNo label
  • Sales targets Q3.xlsxSalesSales teamNo label
  • Pay policy 2025.pdfIntranetEveryoneNo label

3 files open to people who should not see them

Asked by a sales employee

What do people in my team earn?

AI

Based on Salaries 2019 final.xlsx, monthly salaries in your team range from €3,400 to €5,900. The highest is…

Governance

Rules your AI can rely on.

Oversharing is the urgent part. Lasting governance covers six areas, set up with the people who own the data and written down so they can run it themselves.

  • Ownership

    Every site, team and dataset gets a named owner who decides who may use it and answers for its quality.

    You get: An owner list for sites, teams and datasets

  • Classification and labels

    A few sensitivity labels everyone understands, from General to Highly confidential, applied automatically where the content makes it clear.

    You get: A label scheme, with defaults per site

  • Access rights

    Access by group instead of by person, only as wide as the work needs, and checked by the owner on a fixed schedule.

    You get: A group model and recurring access reviews

  • AI policy

    Which AI tools are approved, which data they may use and who decides on a new use case. On one page, in plain language.

    You get: AI usage rules and an approval route

  • Compliance

    GDPR applied to AI: personal data, retention and impact assessments where needed. For the EU AI Act: a list of your AI systems, their risk level and AI training for staff.

    You get: An AI register, ready for your privacy officer

  • Data quality

    An AI answer is only as good as the file it quotes. Outdated versions are archived, definitions agreed and key data tested, so the AI stops citing last year’s figures.

    You get: Agreed definitions and tested key data

Guardrails

The rollout continues safely.

Once the biggest risks are fixed, the rollout picks up again in waves. Guardrails keep it safe as new sites, files and people arrive.

Example: the Copilot rollout, back on track

  1. 1

    Pilot resumesWeek 4

    50 users, on sites that are cleaned up and labelled.

    Gate passed

  2. 2

    Second waveWeek 8

    300 users, after the access review of their sites.

    Gate passed

  3. 3

    EveryoneWeek 12

    The whole company, with the guardrails running.

    Gate passed

  • Safe by default

    New sites start private with an owner and a default label. Open sharing needs a reason.

  • Alerts on new oversharing

    A sensitive file shared with everyone triggers an alert to its owner, not a surprise months later.

  • Access reviews

    Owners confirm who has access each quarter. Leavers and old guests drop off automatically.

  • AI use you can account for

    AI use is logged and checked against the rules, so you can show the auditor what happens with your data.

How we work

From scan to safe rollout.

SolvX starts with a fixed-price oversharing scan. You keep the findings, whether or not we do the rest. If you want, we then fix the biggest risks first and set up governance with your team, so the rollout can go on.

  1. 01

    Weeks 1 and 2

    Scan

    We map what every user, and so every AI assistant, can open, and sort the findings by risk.

  2. 02

    Weeks 3 to 6

    Fix

    Open sharing closed, forgotten sites archived or given an owner, sensitive files labelled. Highest risk first, so the pilot can resume on clean sites.

  3. 03

    From week 7

    Govern

    Owners, labels, access reviews and AI rules agreed with your team, while the rollout widens in waves.

  4. 04

    After go-live

    Keep it safe

    Alerts and reviews run on their own, your team handles them, and we stay on call when you want us.

Questions

About AI and governance.

Does Copilot show files people are not allowed to see?

No. Microsoft 365 Copilot only uses files the user can already open. The problem is that many users can open far more than they should, through old sharing settings, open links and forgotten sites. Copilot makes that visible in seconds.

Should we pause our AI rollout until everything is in order?

Usually not. We fix the highest risks first, keep the riskiest sites out of AI search in the meantime, and let the rollout continue in waves on sites that are clean.

Does this only apply to Microsoft 365 Copilot?

No. Any AI assistant that searches your files with the user’s own permissions, from the AI of your choice to an AI Company Brain, sees only what that user can open. Fixing access once protects every such tool.

Which tools do you use?

Mostly what Microsoft 365 already offers, such as SharePoint sharing reports and Microsoft Purview sensitivity labels and data loss prevention. Some features depend on your licence; the scan shows which ones you have. No extra platform is needed to get started.

What does the EU AI Act ask of us?

For most companies using AI assistants the duties are light: staff who work with AI should know enough to use it responsibly. Stricter rules apply to some uses, for example AI in hiring. To know which rules apply to you, you need a list of the AI systems you use, what for and how risky that use is. We set that up as part of the AI rules.

How is this different from the Data Governance Health Check?

The Health Check is a fixed-scope audit of the governance on your data platform, ending in a gap report. AI & Data Governance makes your documents and data safe for AI, and we build and run the fix with you.

Next step

What can your AI see today?

Tell us where your AI rollout stands, or where it stalled. We show you what a scan would cover and how quickly the rollout can go on.