AI Sovereignty in Software Testing: Keep Control of Your Models, Your Data and Your Tests

AI sovereignty in software testing: your team decides which AI sees its screens and where test data lives, and can change it without rewriting a test.

By Ait Nohaila, Business Development Representative 10 min read

AI sovereignty in software testing means your team decides which AI model sees its work, where its test data is processed, and can change that decision tomorrow without rewriting a single test.

Most teams never made that decision. They chose a testing tool, and the tool chose the rest: which model reads their screens, on whose servers, at what price. For a QA lead or a CTO, that is a governance question hidden inside a tooling choice.

What is AI sovereignty?

Definition

AI sovereignty is a team's power to decide which AI model works on its data, where that work happens, and to change the decision when it wants to.

  • The model: you choose it, and you can replace it.
  • The place: you know where your data is processed, and you can keep it on your own computers.
  • The way out: changing the model or the place does not cost you the work you have already done.

The words sovereign AI are also used for countries and their computing power. This article is about a smaller scale, the one a QA lead can act on this quarter: one team, its tests and its tools.

TermThe question it answers
Data sovereigntyWhere is our data stored, and whose rules apply to it?
AI sovereigntyWhich model works on our data, where does it run, and can we replace it?
AI governanceWho in our company decides those things, and where is it written down?

Why AI sovereignty matters most in software testing

Three things make testing a special case.

  • An end-to-end test goes through the real screens of your app, so it sees what a user sees: customer names, prices, account numbers.
  • It runs on features that are not released yet, the ones you would least like to see outside the company.
  • And it runs on every build, so whatever leaves your company leaves it every day.

An AI feature in a testing tool works on exactly those screens.Where the copies of a test run go, and what privacy rules ask of them, is covered in data privacy in test automation.

Three ways a team loses control of its AI

Nobody signs away control in one decision. It goes in small steps, each of which looked like a convenience.

How control is lostWhat it looks like a year later
Every screen goes to a model you did not chooseCustomer data and unreleased features have left the company on every run, under terms nobody on the team has read.
Tests only run inside someone else's cloudA price change, an outage or a new policy can stop your releases, and leaving means building every test again.
An AI verdict nobody can inspectA release went out because a model said pass, and nobody can show what it looked at or why.
The problem is rarely the choice a tool made. It is that nobody on your team made it.

What each of these costs, and how to check a tool for it, is the subject of AI test automation without vendor lock-in.

The four decisions that should be yours

Sovereignty is practical. It comes down to four decisions, and to who makes each one. GUI Test Studio, which automates end-to-end tests of any app without code, leaves all four with your team.

The decisionWho makes it with GUI Test Studio
Which AI looks at your workYou do. A cloud AI with your own key, a local AI model, ours or your own, or no AI. It is one setting.
Where the test data livesYou do. Tests, recordings and reports are files on your computers, in folders you choose.
Which assistant creates testsYou do. Any assistant that speaks an open standard, a local AI model, ours or your own, or no assistant: you record the test yourself.
Whether a run needs AI at allYou do, and the answer can be no. A recorded test replays by itself.

Which AI: a cloud AI with your own key, a local AI model, or none

AI Smart Verification is the optional second look at a failed check: it gives the final verdict of that check and explains the difference in plain words. It is off until you switch it on, and one settings card says where it runs: at a cloud provider with your own key, or on a local AI model, one we provide or your own.

Your team holds the switches

A cloud AI

With your own key, from the provider you choose

A local AI model

Ours or your own, on your own computer

No AI

Every check decides on its own comparison

Your tests: the same files, whichever switch is on.

Three choices, one setting. Here the team has chosen a local AI model, ours or its own, and it can choose again tomorrow.
Your choiceWhat it means for control
A cloud AI, with your own keyYou pick the provider, OpenAI or Anthropic, and the model. Your key talks to the provider directly, under your own agreement with it.
A local AI modelA model we provide or one you bring runs on your own computer. Nothing of the screen leaves it, and the model changes only when you change it.
No AIEvery check passes or fails on its own comparison. Nothing depends on a model.
One settings card holds the whole decision: on or off, where it runs, which model.

Every option, with what to expect of each, is in the AI Smart Verification guide.

Data sovereignty in test automation: files on your own computers

A test is a file your team owns: its steps, its pictures and its recording. Every run saves its report and its video in a folder next to the test. Nothing of a test is kept on our side, so where it is stored, who can open it and how long it is kept are your decisions.

That is also what makes the results usable as proof later.Test evidence for audits shows what an auditor asks of a test run, and what each run keeps.

Which AI assistant creates your tests

A test can be created from a prompt: you describe the workflow, and an AI assistant performs it on your screen and saves an ordinary test. The connector speaks the Model Context Protocol, an open standard, so the assistant is your choice and not ours.

You have three ways, and you can mix them.

  • Use the assistant your team already works with: Claude Desktop, Claude Code, ChatGPT with Codex or VS Code.
  • Use a local AI model, one we provide or your own, so the request and the screen stay on the computer.
  • Or use no assistant, and record the test by doing the workflow once.

Whichever way a test was created, the result is the same kind of file, and it never needs the assistant again.See Create a test from a prompt, and MCP server for testing for what the open standard changes.

Does a test run need AI at all?

No, and this is the decision that protects the other three. A recorded test replays by itself: the same steps in the same order at every run, with no AI driving the app. The same test on the same app gives the same answer.

Why that matters for trust in a result is the subject of use AI to create the test, not to improvise every run.

Can you inspect what the AI decided?

Choosing the model is half of control. The other half is being able to read what it decided, long after the run.

A verdict nobody can inspect

AI says: pass

What did it look at, and why? Nobody can say.

A verdict saved with the run

What was compared

The expected picture, beside what the run found

The verdict

Failed

The reason, in plain words

The Confirm order button changed from active to disabled.

The instructions it was given

Ignore changes to the font size.

Control over an AI includes being able to read what it decided. Each AI verdict is kept with its run, and can be opened later.

In GUI Test Studio, an AI verdict is kept with its run: the assessment, the explanation, every difference found, and the AI Instructions it was given.Those instructions are yours too.A few plain words on a check, such as "Ignore changes to the font size", tell the AI how your team wants that check judged.

Changing your choice later, without rewriting a test

A decision you cannot take back is not a decision you control. So the real test of sovereignty is what a change of mind costs.

This year

A cloud AI, with your own key

Next year

A local AI model, ours or your own

The tests: not recorded again, not rewritten, not even opened.

The AI choice is a setting beside the tests, never a part of them.
ExampleA security review, six months in

A team starts with a cloud AI and its own key. Six months later, its security review asks that no screen leave the company. The team switches the setting to a local AI model, one we provide or its own.

No test is opened, recorded again or rewritten. The next run explains its failed checks with the new model, in the same form as before.

AI governance in testing: a one-page policy

Governance is sovereignty written down. A QA lead can put it on one page, and review it once a year.

The page answers five questions.

  • 1Which AI models may look at our test screens?
  • 2Where may that processing happen: at a provider we have an agreement with, or only on our own computers?
  • 3Which AI assistants may create tests, and against which environments?
  • 4What is kept of an AI verdict, and for how long?
  • 5What still runs when AI is switched off?

Before you choose a tool, AI testing tools: 7 questions to ask turns the same concerns into questions for a vendor.

What to know before you start

LimitWhat it means
A local AI model needs hardwareA local AI model, ours or your own, is fast with a graphics card. Without one it works, slowly.
A cloud AI sees what you send itWith a cloud AI, the pictures of a failed check go to the provider you chose. A check that passes is never sent.
The choice is set on each computerThe AI setting belongs to each computer, and to the command line for a pipeline run. Write your team's choice down, and set your machines alike.

Key takeaways

  • 1AI sovereignty is a decision you keep: which model, where it runs, and the freedom to change both.
  • 2Testing is where it matters most, because tests see customer data and unreleased features on every build.
  • 3Control is lost in three ways: a model you did not choose, tests that only run in someone else's cloud, an AI verdict nobody can inspect.
  • 4In GUI Test Studio four decisions stay yours: which AI, where the data lives, which assistant, and whether a run needs AI at all.
  • 5A change of mind is a setting: a cloud AI with your own key, a local AI model, ours or your own, or none, and the tests stay as they are.

Questions

What is AI sovereignty?

AI sovereignty means your team decides which AI model works on its data, where that work happens, and can change the decision later. In software testing it covers the model that looks at your screens, the place your tests and results are kept, and the assistant that creates your tests.

What is the difference between AI sovereignty and data sovereignty?

Data sovereignty is about where your data is stored and whose rules apply to it. AI sovereignty adds the model: who chooses it, where it runs, and whether you can replace it. A testing team needs both, because an AI feature sends data to a model every time it is used.

Why does AI sovereignty matter in software testing?

Because tests see your most sensitive screens, every day. An end-to-end test goes through customer records and features that are not released yet. A tool that sends those screens to a model you did not choose takes the decision out of your hands on every run.

Can I use AI in testing without sending my screens to a cloud?

Yes. In GUI Test Studio, AI Smart Verification and test creation from a prompt can both run on a local AI model on your own computer, one we provide or one you bring. Nothing of the screen leaves the computer.

Do my tests need AI to run?

No. In GUI Test Studio a recorded test replays by itself: the same steps in the same order, with no AI driving the app. AI Smart Verification is optional and off until you switch it on, so every test still runs and gives a verdict without any model.

Can I change the AI model later without rewriting my tests?

Yes. The AI choice is a setting, not a part of the test. Move from a cloud AI to a local AI model, ours or your own, or switch AI off, and every test runs as before: nothing is recorded again and nothing is rewritten.

Which AI assistants can create tests in GUI Test Studio?

Claude Desktop, Claude Code, ChatGPT with Codex and VS Code can, through a connector that speaks an open standard, the Model Context Protocol. A local AI model, one we provide or one you bring, can create tests too. Recording a test yourself needs no assistant at all.

What should an AI policy for testing say?

It should answer five questions: which AI models may look at test screens, where that processing may happen, which assistants may create tests, what is kept of an AI verdict, and what still runs when AI is switched off. One page is enough.

Claude, ChatGPT, Codex, OpenAI, Anthropic, Visual Studio Code and Windows are trademarks of their respective owners.

All articles