Almost every AI conversation I have with a security or IT lead in the UAE starts with the same line: our data can't leave the country. It is a fair starting point. It is also where a lot of pilots get stuck, because nobody has pinned down what the line actually means for the task in front of them.

In practice, that one sentence hides three separate questions. Where is the data stored? Where does the model actually run? And who else can see it, keep it or learn from it? A setup can pass the first question and fail the other two. Here is how I would work through them.

Start with what the rule actually is

Before any technology choice, find out whether your rule comes from the law, your regulator or your own policy. They lead to different answers. This is not legal advice, and your lawyer should have the final word, but here is the broad picture as I understand it.

  • The federal Personal Data Protection Law (Federal Decree-Law No. 45 of 2021) does not ban personal data from leaving the UAE. It allows transfers in listed cases, such as to countries with adequate protection, under a contract that holds the receiver to the same standards, or with the person's explicit consent. The Executive Regulations that are meant to spell out the details had not been published when I wrote this, so check the current position before you rely on it.
  • The DIFC and ADGM have their own data protection laws, and the federal law does not apply inside them. If you are based in one of those free zones, read that law instead.
  • Health data is stricter. Federal Law No. 2 of 2019 on technology in health fields says health data related to services provided in the UAE may not be stored, processed or transferred outside the country, except in cases the health authorities allow.
  • Banks, insurers and government entities usually have rules from their own regulators on top of all this.

For a lot of companies, the strictest rule turns out to be their own internal policy. That is fine. It just means the policy owner, not the law, is the person who has to sign off, and you should talk to them early.

Storage and processing are not the same thing

This is the part that catches most teams. Cloud providers now have regions in the UAE, and it is easy to assume that picking a UAE region keeps everything in the UAE. It often doesn't.

Many AI services separate where your data is stored at rest from where the model runs your request. You can keep your files, logs and settings in a UAE region while the request itself is sent to whichever region has spare capacity for that model. The settings usually have names like "global", "data zone" or "regional". Only the regional or in-country option keeps the model's work inside the country, and not every model is offered that way in every region.

So the question to ask is not "do you have a UAE region?" It is "for this exact model, in this exact setup, where does my request get processed?" Get the answer in writing, and check it again when you switch models, because it can be different for each one.

Personal accounts are the bigger leak

While a company debates regions, its staff are often pasting customer emails and contracts into personal chat apps on their phones. That data has left the building in the worst possible way: no contract, no controls and no record.

Business plans from the main AI providers generally come with controls that personal accounts don't have, such as turning off training on your data, setting how long conversations are kept and managing who has access. Some providers also offer storage, and in some cases processing, in the UAE for business customers. Read the terms of the plan you are actually on rather than the marketing page. A sanctioned tool with clear settings is usually safer than a ban that people work around.

What to ask in the contract

Whatever setup you choose, these are the questions I would want answered in writing before any real data goes in:

  • Where is our data stored at rest, and where does the model process our requests?
  • Which other companies (subprocessors) touch our data, and in which countries?
  • How long are prompts, outputs and logs kept, and can we change that?
  • Is our data used to train or improve any model? Can we turn that off by default?
  • Can the provider's staff ever read our data, for example for abuse checks, and under what conditions?
  • How do we delete everything when we leave?

If a vendor can't answer these clearly, that tells you something.

Running a model yourself

There is another option: run an open model on servers you control, either your own or in a UAE cloud region. Then the request never leaves your setup at all.

The trade-off is real. You now run the model, keep it updated and pay for the hardware whether you use it or not. For some tasks an open model does the job well. For others it falls short of the larger hosted models. The only way to know for your work is to test it on your own cases, the same way you would test any other option.

How we handle it in a pilot

When we run a pilot, data location is one of the first things we settle, not the last. We start by sorting your real test cases by how sensitive they are: which contain personal data, which contain health or financial records, and which contain nothing sensitive at all. Then for each task we only test setups that meet the rule for that data, and we write down where the data goes as part of the scorecard, next to accuracy and cost.

Sometimes the answer is to split the work. The sensitive step stays with a person or a plain rule, and AI handles the rest. Sometimes the in-country option is good enough and the question goes away. Either way, your security team sees exactly what was tested and where the data went before anything goes live.

If data location is what's holding up your AI plans, start a chat with us on the site, or book a call through the Applied Labs page. Bring your policy and one task you'd like to automate, and we'll show you which setups would actually meet it.