If you're building an AI-powered product in Uzbekistan, the legal landscape is far more complex than it appears at launch. We break down the key risks — from data processing to liability for 'algorithmic decisions' — and explain what you need to lock down before going live.
The short answer: Uzbekistan doesn't yet have a dedicated AI law — but that doesn't mean your service operates in a legal vacuum. The regulatory framework emerges from the intersection of personal data protection, e-commerce, consumer rights, and general civil liability rules. The sooner you internalize this, the cheaper your compliance journey will be.
Key takeaways:
---
When we were building Pactum, one of the first questions we asked was: "Is there actually any AI regulation here?" The answer is yes — just indirect. Your service collects data → data protection law applies. Your service makes recommendations to users → consumer protection law applies. Your service generates revenue → tax and corporate law apply.
The global trend is moving toward direct AI regulation (the EU AI Act is the clearest example), and Uzbekistan, as it actively integrates into the digital economy, will inevitably follow suit. My strong advice: build your processes now so that tightening regulation doesn't force you to re-architect your product from scratch.
---
Every AI service runs on data. If that data includes names, contact details, behavioral patterns, biometrics, or any other information that can identify an individual, you fall squarely under personal data protection legislation.
What this means in practice:
Practical tip: conduct a data mapping exercise — create a clear picture of what data you collect, where it's stored, and who has access. It takes a few hours and resolves about 80% of regulatory questions before they're ever asked.
---
| Scenario | Who Bears Liability | How to Reduce Risk |
|---|---|---|
| AI gave incorrect advice and the user suffered a loss | The developer / service operator | Disclaimers, liability caps in ToS, framing output as informational |
| AI processed data without consent | The data controller | Explicit consent mechanisms, process audits |
| B2B context: client relied on AI output and made a wrong call | Depends on the contract | Clear liability allocation in the service agreement |
| AI in a regulated sector (healthcare, finance) | The company + potentially the director personally | Licensing, additional compliance procedures |
The algorithm is not a legal entity and cannot be held liable. You can. That's exactly why your Terms of Service and B2B contracts must clearly describe what your AI does — and what it doesn't do — and where the line of responsibility sits.
---
If your AI service touches healthcare, legal services, financial advisory, or education, expect heightened regulatory scrutiny. These sectors frequently carry licensing requirements, and an automated service may fall within their scope regardless of whether you call it a "tool" or a "full advisor."
The core rule: if your product does what a licensed professional does, consult a lawyer before launch — not after your first regulatory complaint.
---
Another gray area: ownership of AI-generated content. Who holds the rights to the text, images, or code your service produces? Traditional copyright law requires a human author. In practice, this means:
---
---
Does an AI service require a special business registration in Uzbekistan?
No dedicated registration category exists for AI services. You register a legal entity or sole proprietorship through the standard process. The exception is if your activity falls within a licensed sector.
Can I store user data on servers outside Uzbekistan?
Yes, but it requires a legal basis. Personal data legislation sets out requirements for cross-border transfers. The specific conditions vary and should be verified against your particular circumstances.
What should my Terms of Service say about AI features?
Describe the functionality honestly, frame recommendations as informational rather than advisory where that accurately reflects the product, and explicitly limit the company's liability for losses arising from automated outputs.
If the AI makes an error and a user suffers a loss — does that mean litigation?
Not necessarily, but a formal claim is a real possibility. A well-drafted ToS and documented processes are your first line of defense. The clearer your product's boundaries are defined, the weaker the grounds for a successful claim.
When will Uzbekistan introduce a dedicated AI law?
No confirmed timeline exists. Regulatory initiatives are under discussion, but the situation is evolving. I recommend tracking developments and building your product on a "compliance by design" basis starting now.
---
*This article is for general informational purposes only and does not constitute individual legal advice. Specific legal questions must be assessed in the context of your particular business — please consult a qualified professional.*
---
If you'd like to work through the legal side of your AI product in detail — book a consultation. At Pactum, this is exactly the kind of case we handle: where technology meets law.

Founder of the Pactum legal platform. Writes about the legal side of IT, AI and startups in Uzbekistan — from data protection and IT Park to venture deals.
Notarial fees in Uzbekistan are made up of two distinct components: a state duty and a notarial tariff for legal and technical work. In practice, the total amount depends on the type of action, the value of the asset involved, and the complexity of the document — let's examine each element in detail.
Notarial certification of the time of document presentation is an official confirmation that a specific document existed and was presented to a notary at a particular moment in time. In this article, I explain when this notarial act is necessary, what to bring with you, and what to keep in mind.
Notarial evidence preservation is an official procedure for recording facts that may disappear before litigation begins. Learn when it is necessary, how the process works, and what documents you will need.