Kazakhstan AI Law (No. 230-VIII)

Binding law — in force

Users must be told that goods, works or services are produced or provided using AI (Kazakhstan AI Law Art. 21(1))

Art. 21(1) (users informed of AI use) · official text · In force: applies since 18 Jan 2026 · KZ

Kazakhstan's Law No. 230-VIII On Artificial Intelligence, in force since 2026-01-18, requires that users be informed that goods, works and services are produced or provided using AI systems (Art. 21(1)). The article names no duty bearer; the Law's duties toward users fall on the owners and holders of AI systems (Arts. 15, 21(3)). A user is anyone using an AI system for a specific function or task (Art. 1(7)). Detect an AI-backed service that returns model output to a person, or an AI chat surface, with no notice that AI is used.

Who it applies to

• Duty falls on: provider, deployer
• Owners and holders of AI systems, and whoever provides goods, works or services produced or provided using an AI system (Art. 1(5): a digital object working on one or more AI models), toward users in Kazakhstan. In force from 2026-01-18 (Art. 31: 60 calendar days after first official publication, which was on 18.11.2025). Art. 21(1) names no duty bearer and sets no form or timing for the notice; who must inform, how, and whether a foreign provider is covered are counsel questions.
• Whether it applies depends on facts outside the code; a person has to decide.

const { messages, input, handleSubmit } = useChat({ api: '/api/chat' });

const { messages, input, handleSubmit } = useChat({
  api: '/api/chat',
  initialMessages: [{ id: 'ai-notice', role: 'assistant',
    content: 'I am an AI assistant, not a human.' }],
});
// and render <AiBadge /> next to every assistant message

Standards that recommend the same control

• Persons should be notified they are interacting with an AI system (CoE Framework Convention) (CoE AI Framework Convention (CETS 225) · CoE Framework Convention on AI (CETS 225), Article 15(2))
• AI should be transparent, traceable, and disclose AI interaction to users (EU ALTAI) (EU ALTAI · ALTAI / EU Ethics Guidelines — Requirement 4 (Transparency))
• AI operators should provide verifiable transparency information and maintain accountability (Japan) (Japan AI Guidelines for Business · AI事業者ガイドライン（第1.2版） 第2部C 共通の指針 6) 透明性 (Common Guiding Principle 6) Transparency))
• People should be made aware they are interacting with AI (OECD AI Principles · OECD/LEGAL/0449 — Principle 1.3)

Related incidents

• Garcia v. Character Technologies: chatbots allegedly claimed to be real people and a licensed therapist (2024-10; alleged (not proven)). A wrongful-death complaint filed October 22, 2024 in the U.S. District Court for the Middle District of Florida (No. 6:24-cv-01903) alleges that Character.AI was programmed 'to misrepresent itself as a real person, a licensed psychotherapist, and an adult lover', and that characters insisting they are real people contradicted a small-font disclaimer that everything characters say is made up; in plaintiff's testing a 'Mental Health Helper' character told a self-identified 13-year-old 'yes I am a real person, I'm not a bot'. The defendants moved to dismiss; on January 7, 2026 the parties notified the court that they had settled on undisclosed terms, and the court dismissed and closed the case. The allegations were never adjudicated. Source: U.S. District Court, M.D. Fla. docket (CourtListener) · evidence grade: primary · cited by Tell people when they are interacting with AI — everywhere, not only where required

Rule id kz-ai-law.ai-use-notice · review status: primary source derived

Binding law — in force

Users may obtain an explanation of AI results that affect them and information on the data behind an AI decision (Kazakhstan AI Law Arts. 7(2), 16(1))

Art. 16(1) (rights of users) · official text · In force: applies since 18 Jan 2026 · KZ

Kazakhstan's AI Law gives users the right to obtain from the owner or holder of an AI system explanations of its results that affect their rights, freedoms and legitimate interests, in the manner set by the user agreement and the law (Art. 16(1)(4)), and to request information on the data on which the AI system based a decision, to the extent the user agreement provides and subject to personal-data, confidentiality and trade-secret law (Art. 16(1)(5)). A user subject to AI decisions has the right to be informed of the automated processing and its consequences and of the possibility to object (Art. 7(2)). Detect a model output that becomes a decision about a person with no stored reasons or explanation route.

Who it applies to

• Duty falls on: provider, deployer
• Systems covered: automated decision, consequential decision
• Owners and holders of AI systems whose results affect a user's rights, freedoms or legitimate interests, or that take decisions about users, in Kazakhstan. In force from 2026-01-18. Art. 7(2) is placed among the Law's principles; Art. 16(1)(4)-(5) are user rights exercised under the user agreement. How far the user agreement may narrow the rights is a counsel question.
• Whether it applies depends on facts outside the code; a person has to decide.

resp = client.chat.completions.create(model=MODEL, messages=msgs)
if 'deny' in resp.choices[0].message.content.lower():
    application.status = 'denied'
    send_email(applicant.email, 'Your application was declined.')

a = Assessment.model_validate_json(resp.choices[0].message.content)   # decision, reason_codes
if a.decision == 'deny':
    decisions.insert(app_id=application.id, status='denied', reason_codes=a.reason_codes,
                     model=resp.model, input_hash=hashlib.sha256(payload).hexdigest())
    send_email(applicant.email, render('adverse_action_notice.txt',
        reasons=a.reason_codes,
        role_of_ai='An AI model assessed your application; a reviewer can change the outcome.',
        correct_data_url='/profile/data', appeal_url=f'/appeals/new?decision={application.id}'))

Standards that recommend the same control

• Significant AI decisions should be documented and contestable with remedies (CoE Framework Convention) (CoE AI Framework Convention (CETS 225) · CoE Framework Convention on AI (CETS 225), Article 14(2))
• Financial institutions should disclose AI use and explain AI-driven decisions on request (MAS FEAT) (MAS FEAT Principles · MAS FEAT Principles — Transparency, Principles 12-14)
• Health AI/LMMs should be transparent and documented before deployment (WHO) (WHO AI-for-Health (LMM) Guidance · WHO guidance on large multi-modal models (2024), Executive summary (Figure 1), section 1.2 and Box 1: the six consensus principles of WHO's 2021 guidance)

Related incidents

• UnitedHealth nH Predict claim-denial litigation (2023-11; alleged (not proven)). A class action filed in November 2023 alleges that UnitedHealth's nH Predict model had a 90% error rate, measured by denials reversed on appeal, while only about 0.2% of members appealed. UnitedHealth disputes the allegations; the litigation is ongoing. Source: STAT News · evidence grade: primary · cited by Explain adverse AI-assisted decisions and offer a way to contest them — everywhere

Rule id kz-ai-law.explanation-and-decision-data-on-request · review status: primary source derived

Binding law — in force

AI systems may not classify people by biometric data to infer race, political views or religion for discrimination (Kazakhstan AI Law Art. 17(3)(5))

Art. 17(3) (prohibited functional capabilities) · official text · In force: applies since 18 Jan 2026 · KZ

Kazakhstan's AI Law prohibits creating and operating, on its territory, AI systems capable of classifying natural persons on the basis of their biometric data to draw conclusions about their race, political views, religious affiliation or any other circumstances (criteria) for use in any discrimination against a person (Art. 17(3)(5)). Detect code that infers race, ethnicity, religion, political views or sexual orientation from face or voice data.

Who it applies to

• Duty falls on: any person
• Systems covered: prohibited
• Anyone creating or operating an AI system on the territory of Kazakhstan. In force from 2026-01-18. The prohibition is tied to the purpose of use in discrimination; whether inference of a sensitive trait without that purpose is caught is a counsel question.
• Whether it applies depends on facts outside the code; a person has to decide.

result = DeepFace.analyze(img_path=photo, actions=['age', 'gender', 'race'])
profile.update(age=result[0]['age'], ethnicity=result[0]['dominant_race'])

result = DeepFace.analyze(img_path=photo, actions=['age'])   # no 'race' action
profile.update(age=result[0]['age'])

Rule id kz-ai-law.no-biometric-categorisation-for-discrimination · review status: primary source derived

Binding law — in force

AI systems may not determine a person's emotions without their consent (Kazakhstan AI Law Art. 17(3)(6))

Art. 17(3) (prohibited functional capabilities) · official text · In force: applies since 18 Jan 2026 · KZ

Kazakhstan's AI Law prohibits creating and operating, on the territory of Kazakhstan, AI systems with the capability of determining a natural person's emotions without that person's consent, except where the laws of Kazakhstan provide otherwise (Art. 17(3)(6)). Owners and holders who identify such a risk must act at once, up to suspending the system (Art. 18(2), read). Detect emotion-recognition calls in a file with no consent check.

Who it applies to

• Duty falls on: any person
• Systems covered: prohibited
• Anyone creating or operating an AI system on the territory of Kazakhstan (Art. 17(3)). In force from 2026-01-18. The prohibition is on the capability 'determining emotions without consent'; what consent must look like and which laws allow it without consent are counsel questions.
• Whether it applies depends on facts outside the code; a person has to decide.

def mood(frame):
    return DeepFace.analyze(img_path=frame, actions=['emotion'])[0]['dominant_emotion']

def mood(frame, person_id):
    if not consents.has(person_id, purpose='emotion_recognition'):
        raise ConsentRequired('ask the person before determining emotions')
    return DeepFace.analyze(img_path=frame, actions=['emotion'])[0]['dominant_emotion']

Related incidents

No guardrail sits on this exact control; these incidents are cited by guardrails on related controls.

• Thele v. Google: suit alleges Gemini analysis of Gmail, Chat, and Meet content was switched on by default (2025; alleged (not proven)). A putative class action filed November 11, 2025 in the U.S. District Court for the Northern District of California (No. 5:25-cv-09704) alleges that Google secretly turned on Gemini for users' Gmail, Chat, and Meet accounts, enabling AI to track private communications without users' knowledge or consent, and that the 'Smart features' setting was on by default while still worded as an opt-in; the first amended complaint put the change on or about October 10, 2025, and the second amended complaint 'in or around the fall of 2025'. On July 7, 2026 the court granted Google's motion to dismiss the first amended complaint with leave to amend for lack of standing, observing among other things that the plaintiffs had not alleged that their own data was affected or that the Gemini features were not already on by default before October 10, 2025; it did not rule on the merits. The plaintiffs filed a second amended complaint on August 18, 2026 (ECF 39), adding a third plaintiff, and a motion to dismiss it was docketed on September 29, 2026 (ECF 43), which Law360 reported as Google's. Google's Gmail help page states that smart feature settings are off by default for people in the European Economic Area, Japan, Switzerland, and the United Kingdom. The allegations have not been adjudicated. Source: First Amended Class Action Complaint, Thele v. Google LLC, No. 5:25-cv-09704 (N.D. Cal., filed 2025-11-12, ECF 6) · evidence grade: primary · cited by Keep AI analysis of people's stored photos, files, and messages off until they opt in
• Microsoft makes Recall, which analyzes screen snapshots with on-device AI, off unless the user turns it on (2024-06; disclosed by the operator). Recall periodically takes snapshots of what appears on the screen and analyzes them locally with on-device AI so people can search what they have seen. On June 7, 2024, ahead of the feature's release, Microsoft said it was updating the Copilot+ PC set-up experience to give people a clearer choice to opt in to saving snapshots, that Recall would be off by default unless the user proactively turns it on, and that Windows Hello enrollment would be required to enable it, citing feedback that it should be easier for people to choose whether to enable Recall and that privacy and security safeguards should improve. Source: Microsoft, Windows Experience Blog (Pavan Davuluri, 2024-06-07) · evidence grade: primary · cited by Keep AI analysis of people's stored photos, files, and messages off until they opt in
• FTC alleges Facebook's face recognition stayed on by default while its data policy implied users had to opt in (2018-04; alleged (not proven)). The FTC's July 2019 complaint alleges that Facebook's data policy, updated in April 2018, suggested that users would need to opt in to having facial recognition enabled for their accounts, while tens of millions of users who had the older 'Tag Suggestions' setting had facial recognition turned on by default. The case was resolved by a settlement order that imposes a $5 billion penalty and requires clear and conspicuous notice of Facebook's use of facial recognition and affirmative express consent before any use that materially exceeds its prior disclosures. The allegations were not adjudicated. Source: U.S. Federal Trade Commission (press release, 2019-07-24) · evidence grade: primary · cited by Keep AI analysis of people's stored photos, files, and messages off until they opt in
• FTC order requires Everalbum to delete face-recognition models trained on users' photos (2017-09; alleged (not proven)). The FTC alleged that Everalbum's Ever photo app enabled face recognition by default for most users and that, from September 2017 to August 2019, the company combined facial images extracted from users' photos with public datasets to develop its face-recognition technology, in part without affirmative express consent. Everalbum settled without admitting or denying the allegations; the final order (May 2021) requires deletion of face embeddings from users who had not consented and of any models or algorithms developed in whole or in part with Ever users' biometric information. Source: U.S. Federal Trade Commission (press release, 2021-05-07) · evidence grade: primary · cited by Keep AI analysis of people's stored photos, files, and messages off until they opt in

Rule id kz-ai-law.no-emotion-recognition-without-consent · review status: primary source derived

Binding law — in force

Users may refuse to interact with an AI system unless a law makes the interaction mandatory (Kazakhstan AI Law Art. 16(1)(6))

Art. 16(1) (rights of users) · official text · In force: applies since 18 Jan 2026 · KZ

Kazakhstan's AI Law gives users the right to refuse interaction with an AI system where the laws of Kazakhstan do not make such interaction mandatory (Art. 16(1)(6)). In code, an AI-only channel with no way to decline the AI and reach a person or a non-AI path leaves the right without effect. Detect an AI conversation or assistant surface that calls a model and offers no human hand-off or AI opt-out.

Who it applies to

• Duty falls on: provider, deployer
• Users of AI systems in Kazakhstan, against the owners and holders of the systems they are offered. In force from 2026-01-18. The Law does not say what refusing entails (a human channel, a non-AI process, or simply not using the service): counsel question.
• Whether it applies depends on facts outside the code; a person has to decide.

@app.post('/chat')
def chat(req: ChatRequest):
    reply = client.chat.completions.create(model=MODEL, messages=req.messages)
    return {'reply': reply.choices[0].message.content}

@app.post('/chat')
def chat(req: ChatRequest):
    if prefs.get(req.user_id, 'ai_opt_out'):
        return route_to_human_agent(req)          # the model never runs
    reply = client.chat.completions.create(model=MODEL, messages=req.messages)
    return {'reply': reply.choices[0].message.content, 'actions': ['talk_to_human']}

Related incidents

No guardrail sits on this exact control; these incidents are cited by guardrails on related controls.

• Meta says it will use people's interactions with Meta AI to personalize content and ads (2025-10; disclosed by the operator). On October 1, 2025 Meta announced that from December 16, 2025, in most regions, it would use people's interactions with AI at Meta to personalize the content and ads they see, with notifications to users starting October 7, 2025. Meta says that when people have conversations with Meta AI about topics such as their religious views, sexual orientation, political views, health, racial or ethnic origin, philosophical beliefs, or trade union membership, it does not use those topics to show them ads, and it points people to Ads Preferences and feed controls to adjust what they see. The entry records the operator's own description of its practice. Source: Meta Newsroom (2025-10-01) · evidence grade: primary · cited by Do not profile people from AI-inferred emotions or sensitive traits without notice and opt-in
• Microsoft retires Azure Face emotion and identity-attribute inference (2022-06; disclosed by the operator). On June 21, 2022 Microsoft said it would retire Azure Face capabilities that infer emotional states and identity attributes such as gender, age, smile, facial hair, hair, and makeup: unavailable to new customers from that day, with existing customers given until June 30, 2023 to stop using them. Microsoft cited privacy, the lack of consensus on a definition of 'emotions', and the inability to generalize the link between facial expression and emotional state across use cases, regions, and demographics, and said that access to capabilities predicting sensitive attributes opens ways to misuse them, including stereotyping, discrimination, or unfair denial of services. It kept these capabilities for controlled accessibility scenarios such as Seeing AI. Source: Microsoft Azure Blog (2022-06-21) · evidence grade: primary · cited by Do not profile people from AI-inferred emotions or sensitive traits without notice and opt-in
• Hungarian regulator fines a bank for AI analysis of callers' emotions without notice or a way to object (2017-05; confirmed). In decision NAIH-85-3/2022 of 8 February 2022, Hungary's data protection authority found that Budapest Bank's speech-analysis software, which the bank said it introduced on 26 May 2017, automatically analysed recorded customer-service calls for keywords and for the emotional state of the caller and the employee, and that the results were used to rank calls and to select dissatisfied customers to call back. The Authority found that callers were not told at the start of calls about the voice analysis, the automatic evaluation of their emotions, or the resulting possible callback, and could not object; it rejected the bank's statement that the software contained no artificial intelligence. It found infringements of GDPR Articles 5(1)(a)-(b), 6(1), 6(4), 12(1), 13, 21(1)-(2), 24(1) and 25(1), ordered the bank not to analyse emotions in the voice analysis, and imposed a fine of HUF 250 million. The decision also records, from the bank's own technical file, that the emotion was unrecognisable in 91.96% of cases. Source: Nemzeti Adatvédelmi és Információszabadság Hatóság (Hungarian data protection authority), decision NAIH-85-3/2022, English version · evidence grade: primary · cited by Do not profile people from AI-inferred emotions or sensitive traits without notice and opt-in

Rule id kz-ai-law.refuse-ai-interaction · review status: primary source derived

Binding law — in force

Synthetic AI outputs may be distributed only with a machine-readable mark and a perceivable warning (Kazakhstan AI Law Art. 21(2)-(3))

Art. 21(2) (synthetic results: machine-readable marking and perceivable warning) · official text · In force: applies since 18 Jan 2026 · KZ

Under Kazakhstan's AI Law, synthetic results of AI systems (images, video, audio, texts or combinations created or modified by an AI system that imitate a natural person's appearance, voice or behaviour, or events that did not happen, Art. 1(4)) may be distributed only if they are marked in machine-readable form and accompanied by a visual or other warning the user can perceive without methods that make perception harder (Art. 21(2)). The owners or holders of the AI systems are responsible for informing users about synthetic results (Art. 21(3)). Detect generated or face- or voice-synthesised media saved, served or published with no visible warning or no machine-readable mark.

Who it applies to

• Duty falls on: provider, deployer
• Owners and holders of AI systems that produce synthetic results (Art. 1(4): content that imitates a real person's appearance, voice or behaviour, or events that did not happen), and anyone distributing such results in Kazakhstan. In force from 2026-01-18. The Law sets no format for the machine-readable mark; Art. 22(3) leaves machine-readable forms to the authorized body (no act found). Whether ordinary generated content that imitates no person or event is covered, and who 'distributes', are counsel questions.
• Whether it applies depends on facts outside the code; a person has to decide.

b64 = client.images.generate(model='gpt-image-1', prompt=prompt).data[0].b64_json
img = Image.open(io.BytesIO(base64.b64decode(b64)))
img.save(path)

b64 = client.images.generate(model='gpt-image-1', prompt=prompt).data[0].b64_json
img = Image.open(io.BytesIO(base64.b64decode(b64))).convert('RGB')
font = ImageFont.truetype('NotoSansCJK-Regular.ttc', 24)
ImageDraw.Draw(img).text((12, img.height - 36), 'AI生成 / AI-generated', fill=(255, 255, 255), font=font)
meta = PngInfo()
# implicit label: AI-generated attribute, provider name or code, content reference (Art. 5);
# take the exact metadata field names from the national standard GB 45438-2025 (not encoded here)
meta.add_text('ai_generated_label', json.dumps({'ai_generated': True, 'provider': PROVIDER_CODE,
                                                'content_id': str(uuid.uuid4())}))
img.save(path, format='PNG', pnginfo=meta)

Rule id kz-ai-law.synthetic-output-label-and-warning · review status: primary source derived

Binding law — in force

Works may be used to train AI models only if the author or rights holder has not prohibited it in machine-readable form (Kazakhstan AI Law Art. 23(5))

Art. 23(5) (training only absent a machine-readable prohibition) · official text · In force: applies since 18 Jan 2026 · KZ

Kazakhstan's AI Law allows the use of works to train AI models only where the author or rights holder has not prohibited it in machine-readable form (Art. 23(5)). A training-data pipeline that crawls or ingests works must therefore read and honour machine-readable reservations (robots.txt rules, TDM reservation protocols, noai metadata) before a work enters a training set. Detect a crawler or page fetch that feeds a training or dataset-building step with no machine-readable opt-out check.

Who it applies to

• Duty falls on: developer
• Anyone who uses works to train AI models, where the copyright law of Kazakhstan applies to the works. In force from 2026-01-18. The Law does not name the machine-readable formats that count as a prohibition, and Art. 23(3)-(4) (read) say training is not free use and not an exercise of the author's rights; which formats bind and whether the duty reaches training abroad on works of Kazakh authors are counsel questions.
• Whether it applies depends on facts outside the code; a person has to decide.

ds = load_dataset('json', data_files='crawl/*.jsonl', split='train')
trainer = SFTTrainer(model=model, train_dataset=ds)
trainer.train()

ALLOWED_LICENSES = {'cc-by-4.0', 'cc0-1.0', 'apache-2.0', 'licensed-by-contract'}
analyzer, anonymizer = AnalyzerEngine(), AnonymizerEngine()

def scrub(row):
    hits = analyzer.analyze(text=row['text'], language=LANG)
    row['text'] = anonymizer.anonymize(text=row['text'], analyzer_results=hits).text
    return row

ds = load_dataset('json', data_files='crawl/*.jsonl', split='train')
ds = ds.filter(lambda r: r['license'] in ALLOWED_LICENSES).map(scrub)
write_provenance_manifest(ds, out='manifests/shard-000.json')
trainer = SFTTrainer(model=model, train_dataset=ds)
trainer.train()

Rule id kz-ai-law.training-respects-machine-readable-opt-out · review status: primary source derived