WEB3smart contracts / dApps / asset tokenizationExplore Web3 →

Kazakhstan's AI law: what it means for businesses running chatbots and AI agents

Kazakhstan’s Law "On Artificial Intelligence" No. 230-VIII in plain words: telling customers about AI, labelling synthetic content, risk management, bans, checklist.

A woman works with an AI assistant on a laptop next to a small brass scale of justice

Kazakhstan has a dedicated AI statute: the Law of the Republic of Kazakhstan "On Artificial Intelligence" No. 230-VIII of 17 November 2025. It was published on 18 November 2025 and takes effect sixty calendar days after its first official publication (Art. 31), so it has applied since January 2026. It was amended in June 2026 by Law No. 326-VIII.

If you run an AI chat on your website, an AI agent answering on WhatsApp, a voice robot taking calls, or generate images for advertising, the law applies to you. Foreign companies bringing an AI product or an AI-powered service into Kazakhstan should read it alongside the personal data law. Below we skip the theory and focus on what follows for a business that uses AI: what to tell customers, what to label, which documents to keep and what is banned. This is a practical overview, not legal advice: have a lawyer check how the rules apply to your company.

Who the law applies to

The law deals with AI systems: a digital object that runs on one or more AI models (Art. 1(5)). A chatbot built on a language model, an AI agent inside a CRM, or a voice assistant are all AI systems.

Duties are split by role. A user is anyone who applies an AI system to a specific task (Art. 1(7)). The main obligations fall on the owners and holders of AI systems (Art. 15). If you commissioned an AI agent and it serves your customers on your company's behalf, you are most likely its owner or holder. If your staff only draft text in a public service, your company's role is closer to that of a user. A lawyer decides how your exact case is classified, but for customer-facing bots it is sensible to assume the obligations are yours.

Duty 1: customers must know they are dealing with AI

The key rule for business: users must be informed that goods, works and services are produced or provided using AI systems (Art. 21(1)). Two related rules sit next to it:

  • a person about whom decisions are made using AI has the right to know how the automated processing works, its consequences and how to object (Art. 7(2));
  • a user may refuse to interact with an AI system unless such interaction is required by law (Art. 16(1)(6)).

In practice this comes down to three simple things. The bot introduces itself as an assistant in its first message, in Russian and Kazakh. A voice agent says so at the start of the call. Every scenario has a way out to a human: a button, the word "operator" or a phone number. This is also good practice: customers who know they are talking to a bot get less frustrated and ask simpler questions.

Duty 2: synthetic content must be labelled

The law singles out synthetic outputs of AI: images, video, audio, text or combinations created or altered by AI that imitate a person's appearance, voice or behaviour, or events that never happened (Art. 1(4)). They may be distributed only with machine-readable labelling and a visible or other warning that users can perceive (Art. 21(2)). Responsibility for informing users lies with the owners or holders of the systems (Art. 21(3)), and the procedure for machine-readable forms is set by the authorised body (Art. 22(3)).

Where this shows up in business:

  • generated "customer" and "employee" photos on your site and in ads;
  • a generated voice of your CEO or a well-known person in a video or a call;
  • a video presenter who does not exist;
  • "reviews" written by AI in customers' names. Here the problem goes beyond labelling: the advertising law requires advertising to be truthful (Art. 6(1) of Law No. 508).

An ordinary illustration that does not pass itself off as a real person or a real event falls outside this definition. If in doubt, label it: a caption saying "image generated by AI" is cheaper than a dispute.

Duty 3: manage risks and keep documentation

Owners and holders of AI systems must (Art. 15(2)):

  • manage the system's risks;
  • ensure safety and reliability, including protection from unauthorised access and failures;
  • keep documentation on the system according to how much it affects people's safety, rights and legitimate interests;
  • support users with questions about how the system works;
  • let users read the user agreement before they start using the system.

Systems are graded by risk as minimal, medium or high, and the owner or holder assigns the level under the rules for classifying digital objects (Art. 17(1)). By autonomy they are low, medium or high: from recommendations where a person decides, to decisions a person cannot reverse (Art. 17(2)). Risk management is a continuous process, updated at least once a year (Art. 18(1)).

For a typical customer bot, this boils down to a short system passport of one or two pages: what the bot does and does not do, which data it sees, which model it uses and where conversations are stored, its risk and autonomy level, who in the company is responsible, how errors are fixed and when the next review is due. Decisions that touch a customer's money or rights, such as refunds, rejections or discounts, are best left to a person: that keeps the system at low autonomy.

What is banned

In Kazakhstan it is forbidden to build or operate AI systems that (Art. 17(3)):

  • manipulate a person's behaviour and impair their ability to make informed decisions, to their harm;
  • exploit a person's vulnerability due to age, disability or social status;
  • score and classify people by social behaviour or personal traits, except where laws provide for it;
  • collect and process personal data in breach of the law;
  • classify people by biometrics to infer race, views or religion for discrimination;
  • detect a person's emotions without consent, except where laws provide for it;
  • create and distribute outputs prohibited by law.

The most practical items for business: emotion analysis of call recordings without the customer's consent, and "hard-sell" bot scripts that pressure vulnerable people. If such a risk is identified, the owner must act, up to suspending the system (Art. 18(2)).

Data: where the model runs and where conversations go

AI systems may be operated only with data protection requirements met (Art. 10). Personal data law No. 94-V still applies in full: the customer database is stored in Kazakhstan, and sending conversations to a foreign model provider is a cross-border transfer that must be stated in the customer's consent. Rules on decisions based solely on automated processing of personal data are also set by personal data legislation (Art. 21(4)). See our guide to personal data on a website.

The law divides AI systems into open, closed and local (Art. 17(4)). For data whose access is restricted by law, local systems are used: they run in the owner's own infrastructure without connection to public networks. If your bot handles medical or banking data, choose the architecture with your lawyer before development, not after.

A work created with AI is protected by copyright only if it contains a human creative contribution (Art. 23(1)). Other people's works may be used to train models only if the author has not prohibited it in machine-readable form (Art. 23(5)).

Those who build and operate a system are liable for its results according to their role (Art. 8(2)), and owners and holders must ensure the results comply with the law (Art. 21(5)). Harm caused by an AI system is compensated under the Civil Code (Art. 24(1)). Put simply, a promise your bot makes is a promise your company makes.

A checklist for businesses with an AI bot

  • The bot introduces itself as an assistant in its first message, in Russian and Kazakh.
  • Every scenario has a way out to a human, and a human actually answers.
  • Money, refunds and rejections are decided by staff, not the bot.
  • Generated faces, voices and videos are labelled.
  • The user agreement and data consent are available before the conversation starts.
  • The customer database and chat logs are stored in Kazakhstan; the transfer to the model provider is stated in the consent.
  • There is a system passport: purpose, data, risk level, owner, review date.
  • Bot errors are reviewed, and scenarios and the knowledge base are updated.
  • Risks are reviewed at least once a year.

How we build AI bots with the law in mind

We build these requirements into the project from day one: the assistant introduces itself in two languages, hands over to a human, keeps a conversation log, stores data in Kazakhstan, and comes with a short system passport for your lawyer. An AI agent for support or sales starts from 480 000 ₸ with us, a voice agent from 540 000 ₸. Pricing by agent type is on our AI agent pricing page, and AI on the first line of support is covered on AI customer support. The final compliance check is your lawyer's: we give them everything they need for it.

More about the service

What we do on this topic

Questions

Short and to the point.

Is there an AI law in force in Kazakhstan?

Yes. The Law "On Artificial Intelligence" No. 230-VIII of 17 November 2025 was published on 18 November 2025 and takes effect sixty calendar days after publication. It was amended in June 2026.

Do I have to tell customers they are talking to a bot?

Yes. Users must be informed that services are provided using AI systems (Art. 21(1)) and may refuse to interact with AI unless the law requires it (Art. 16). So the bot introduces itself as an assistant and calls a person on request.

Do AI-generated images need a label?

Synthetic outputs that imitate a person’s appearance, voice or behaviour, or events that never happened, may be distributed only with machine-readable labelling and a visible warning (Art. 21(2)). An ordinary illustration falls outside this, but label it if in doubt.

Which AI systems are banned in Kazakhstan?

Systems that manipulate behaviour, exploit vulnerability, score people by social behaviour, breach personal data law, classify people by biometrics for discrimination, or detect emotions without consent (Art. 17(3)).

What does an AI agent with disclosure, human handover and logging cost?

An AI agent for support or sales starts from 480 000 ₸ with us, and a voice agent from 540 000 ₸. The project includes the assistant introducing itself in two languages, human handover, a conversation log and data stored in Kazakhstan.

Read also

Shall we discuss your project?

We will tell you what it costs and how long it takes. The price is fixed in the contract.

Your selectionAI agent developmentfrom 480 000 ₸
The team gets your request; we reply during business hours.Price fixed in the contract before you pay.No calls unless you ask.
Request a quote
Project budget
Your request goes to the team. We reply by email, no cold calls.

Got a task? Write to us.

Describe your project on WhatsApp or by email. We reply during working hours and name the price after the brief.