Eurosoft

AI and personal data in Uzbekistan: what the 2026 law requires

In January 2026 Uzbekistan amended its law on informatisation to regulate artificial intelligence for the first time. If you use AI on customer or citizen data, two things changed: a machine decision can no longer stand alone where a person's rights are at stake, and mishandling personal data through AI now carries penalties. Here is what the law says and what it means for a project.

Updated · By

What the law introduced

  • Law No. O'RQ-1115 was signed and entered into force on 21 January 2026. It amends the law “On Informatization” and adds a definition of artificial intelligence to Uzbek legislation.
  • The key rule is in the new article 71: a decision that affects a person's rights or freedoms cannot be based only on the conclusions of an AI system. A human must be able to review it.
  • Separate amendments to administrative liability cover the unlawful processing of personal data with AI tools and its distribution through media, telecom networks or the internet. The penalty includes a fine and confiscation of the equipment used.

What it means in practice

  • An AI system may recommend, score, draft and route. A person has to make or confirm the decision that affects someone: a loan refusal, a benefit, a penalty, a medical triage, a hiring decision.
  • That does not make AI useless. It means the workflow needs a review step with a named person, and a log of who confirmed what and when.
  • Chatbots and internal assistants are affected too if they process personal data. A chatbot that answers from a document library is low risk. One that decides who gets a discount or a refusal is where the rule bites.

Where personal data can be stored

  • Uzbekistan's personal data law (No. O'RQ-547, 2019) long required local storage of citizens' personal data. Amendments in March 2026 were reported by law firms as narrowing that requirement to a smaller set of data, biometric and genetic among them.
  • Sources describe the change differently, so check the current text on lex.uz or with a lawyer before you design the infrastructure.
  • Whatever the localisation rule says, storing data on servers you control inside the country is the simplest answer to a client, an auditor or a regulator, and it is what most government-related projects require anyway.

How a compliant private AI is built

  • Documents are indexed and the model runs on the client's own servers, so no request leaves the building.
  • Every answer shows its source, so a person can check it in seconds instead of trusting a black box.
  • Actions with consequences for a person go through an approval step that is logged.
  • Access follows the company's roles: a department sees only the documents it should. This is how Eurosoft's Enterprise RAG Platform is built, and it is why on-premise deployment is a feature, not a limitation.

What to ask a vendor now

  • Where does the model run, and where is the data stored?
  • Can a person review and override every decision that affects a customer?
  • Is there an audit log of who approved what?
  • How are personal data removed when a customer asks?
  • Which provider is behind the API, and what happens to the data sent to it? If the answer is a third-party API abroad, ask for the contract terms on retention and training.

Questions and answers

Does the 2026 AI law ban automated decisions?
It bans decisions that affect a person's rights being based only on an AI system's conclusions. A person must be able to review. Systems that recommend, sort or draft inside a human-controlled process remain fine.
Do we have to keep all personal data in Uzbekistan?
The strict rule was narrowed in March 2026 according to law-firm summaries, but the details differ between sources. Confirm the current position on lex.uz; in doubt, keep the data in the country, which is also the fastest answer to a client's security questionnaire.
What is the penalty for mishandling personal data with AI?
Administrative liability with a fine of 50 to 100 times the minimum wage, up to 41.2 million soums at the 2026 rate, plus confiscation of the equipment used in the offence.
We already use ChatGPT or a cloud API. Is that a problem?
Not automatically, if you do not send personal data and the provider's terms on storage and training are acceptable. For government work and any process that touches citizens' data, a private model on your own servers removes the question entirely.
Who can help set up a private AI in Uzbekistan?
Eurosoft in Tashkent builds private RAG platforms and AI assistants that run on the client's own servers, with source links and role-based access. It is a resident of IT Park Uzbekistan.

See also: services/ai-assistants-and-private-ai · guides/shared-cloud-ai-vs-private-ai

Have a project in mind?

Tell us what you are building and where it has to run. We will come back with a proposed next step.