Central Asia AI Education Audit 2026

Audit date: 17 October 2026. All source access dates are recorded per claim. This is a snapshot; universities are publishing AI guidance monthly.

Executive summary

We audited 31 leading universities across Kazakhstan, Kyrgyzstan, Uzbekistan, Tajikistan, and Turkmenistan, asking two questions of each: (a) does it publish a public, university-level AI usage policy for students and teaching staff (academic integrity or teaching guidance that mentions generative AI)? and (b) what AI-related degree programs or courses does it offer?

Headline results:

• 6 of 31 universities (19%) publish a verifiable, public AI usage policy on their official domain: Nazarbayev University, Al-Farabi Kazakh National University, Toraighyrov University, the American University of Central Asia, the International University of Central Asia (Tokmok), and Westminster International University in Tashkent.
• 18 of 31 (58%) offer a verifiable AI-named degree program or AI course portfolio on their official domain — institutions are teaching AI roughly three times faster than they are governing its use.
• No university in the sample imposes a campus-wide AI ban. Every verified policy follows the global norm of delegated permission plus mandatory disclosure: instructors decide per course, students disclose AI use.
• Kazakhstan dominates the policy landscape: 4 of the 6 confirmed policies are Kazakh, backed by a 2024 inter-university AI standard and an explicit 2025 ministry position against AI bans. Kazakhstan also hosts Qazaq AI Research University (QAIRU), Central Asia's first purpose-built AI university, which enrolled its first 528 students in September 2026.
• Turkmenistan and Tajikistan are near-complete blind spots: 0 of 7 sampled universities in the two countries publish a verifiable AI usage policy. For Turkmenistan this reflects a closed institutional web; Turkmenistan's Ministry of Education met with OpenAI in May 2026 to discuss ChatGPT Edu pilots, but no university-level public policy is visible.
• The audit's UNVERIFIABLE rate (81%) is the highest of any region ACLAS Research has measured — a finding about transparency and discoverability as much as about policy adoption.

Methodology

Sample selection

The sample of 31 universities was constructed from two defensible criteria, stated here explicitly:

1. National flagships and highest-ranked institutions in each country (selected by national prominence and Webometrics/regional ranking position — the institutions a policy survey must not miss): Nazarbayev University, Al-Farabi Kazakh National University, L.N. Gumilyov Eurasian National University, Satbayev University, the Kyrgyz National University (Balasagyn), Kyrgyz State Technical University (Razzakov), Kyrgyz-Turkish Manas University, Osh State University, the National University of Uzbekistan (Mirzo Ulugbek), Samarkand State University, Tashkent University of Information Technologies, Tajik National University, Tajik Technical University (Osimi), the Russian-Tajik Slavonic University, Magtymguly Turkmen State University.
2. IT-specialized and international-format institutions (selected because AI policy is most likely to originate where AI is taught): Astana IT University, IITU, KBTU, SDU, QAIRU, Toraighyrov University (PhD research regulation), the American University of Central Asia, Ala-Too International University, the International University of Central Asia, Westminster International University in Tashkent, INHA University in Tashkent, Turin Polytechnic University in Tashkent, Tashkent State University of Law. Khujand-track schools were deprioritized where web presence was unverifiable, and the Turkmen sample was set to the three internationally visible institutions (IUHD, Oguz Han Engineering and Technology University, Magtymguly TSU).

Country split: Kazakhstan 10, Kyrgyzstan 7, Uzbekistan 7, Tajikistan 3, Turkmenistan 4.

Verification protocol

The protocol is copied from ACLAS Research's Global University AI Policy Audit (14 October 2026):

• Every university is classified confirmed-YES / confirmed-NO / UNVERIFIABLE on the policy question.
• YES requires a policy document on the university's official domain — a regulation, academic policy, senate-approved policy, or strategy with substantive teaching/academic-integrity AI provisions. URL and access date are recorded for every YES.
• UNVERIFIABLE means "could not verify" — dead links, portal-gated intranets, PDFs that fail to fetch, and pages in languages the auditor cannot read — never "confirmed absent."
• Secondary-literature claims that cannot be verified on an official domain are excluded (the reference audit caught two such errors; this audit applied the same filter — e.g., press reports about Ala-Too's AI programs and TimesCA's report on Al-Farabi's academic policy were treated as leads, not evidence, until official-domain documents were found).
• Searches were conducted in Russian, Kazakh, Uzbek, and English (e.g., "политика использования ИИ", "жасанды интеллект саясати", "Положение о применении технологий искусственного интеллекта"). Where a page was in a language the auditor could not read, the result was marked UNVERIFIABLE rather than NO.
• A confirmed-NO was only recorded where the auditor actively inspected the relevant code of conduct or academic policy page and found no AI provision. All other non-YES cases are UNVERIFIABLE; true adoption is therefore likely somewhat higher than reported.

Audit date

17 October 2026. Universities are publishing AI guidance continuously; treat every number below as a snapshot.

Data table 1: AI usage policies

"YES" = a public AI usage policy verified on the university's official domain. See the notes column for scope caveats — two YES cases are scope-limited (doctoral research; institutional strategy), which is stated rather than hidden.

Policy totals: 6 YES (19.4%), 0 confirmed NO, 25 UNVERIFIABLE (80.6%).

Data table 2: AI-related degree programs and courses

"Verified" = program, faculty, or course listing on the university's official domain (URL + access date). Degree titles use the university's own published naming and national qualification codes where shown.

Program totals: 18 of 31 universities (58%) have a verifiable AI-named degree program, AI faculty, or AI course portfolio on their official domain.

Key findings

1. Central Asia records the lowest verified policy-adoption rate ACLAS Research has measured: 19%.

Six of 31 sampled universities publish a public AI usage policy on their official domain. This is far below the 92% measured among the QS top-50 and below the 46% among Global South flagships in the companion global audit (14 October 2026). The gap is not only about adoption: the UNVERIFIABLE rate here (81%) is the highest of any region audited, meaning discoverability and transparency are themselves part of the finding.

2. Kazakhstan is the region's policy hub — and the mechanism is visible.

Four of the six confirmed policies are Kazakh. Kazakhstan is the only Central Asian country with a documented national pipeline converting strategy into institutional regulation: a 2024 inter-university standard for AI use in higher education, a 2025 Ministry of Science and Higher Education position against campus AI bans, the AI-Sana national program, and ministry procurement of AI-detection tools via Antiplagiat.Kazakhstan (TimesCA, May 2026). National direction is measurable in institutional output.

3. Delegation won in Central Asia too.

No sampled university imposes a campus-wide AI ban — consistent with the global finding (zero bans among the top 100). Every verified policy uses the delegated permission + mandatory disclosure model: NU and AUCA delegate per-course rules to instructors (AUCA mandates an AI statement in every syllabus); Al-Farabi, IUCA, and Toraighyrov require students to disclose the tool, version, and stage of AI use, with responsibility for AI-generated errors resting on the author. IUCA's policy even allows instructors to mandate prompt submission for grading.

4. The teach–govern gap: AI degrees outrun AI rules by 3 to 1.

Eighteen of 31 universities (58%) offer a verified AI-named degree program, AI faculty, or AI course portfolio, but only six (19%) publish usage guidance. Kazakhstan opened a purpose-built AI university (QAIRU, first intake 528 students, September 2026); Samarkand State built a Faculty of Artificial Intelligence and Digital Technologies; ENU runs a 72-hour professional AI/robotics program. Institutions are minting AI credentials far faster than they are writing the rules for using AI in their own classrooms.

5. The earliest movers are the region's international-format universities.

Nazarbayev University approved its teaching policy on 18 October 2023 — among the earliest institutional AI policies anywhere in the post-Soviet space. AUCA followed on 11 January 2024. Both are English-medium, internationally partnered institutions (NU was also entry #31 in the 14 October 2026 global audit). Al-Farabi's 2024 regulation and WIUT's 2025 strategy then show the practice spreading to national flagships and private international-format universities — a diffusion pattern from international institutions inward.

6. Turkmenistan is a policy blind spot despite visible state-level AI engagement.

Zero of four Turkmen institutions in the sample publish a verifiable AI usage policy. Yet the state is active: in May 2026 Turkmenistan's Ministry of Education met OpenAI at the Education World Forum to discuss ChatGPT Edu pilots (TimesCA / Golden Age); in June 2026 the ministry issued nationwide mobile-device rules (Order No. 197); the OSCE trained Turkmen IT students in responsible AI use (October 2025). The pattern is top-down modernization with no publicly visible university-level AI governance — opacity, not absence of activity, is the finding.

7. Tajikistan shows a strategy–institution gap.

Tajikistan adopted an AI strategy running to 2040 and has announced AI bachelor/master specialties at technical universities, but none of the three Tajik universities in the sample publishes a verifiable AI usage policy or a verifiable AI program page on its official domain. National strategy has not yet produced visible institutional instruments — a lag worth re-auditing in 2027.

8. Kyrgyzstan: a strong lead, a thin middle.

Kyrgyzstan produced two of the six policies (AUCA, IUCA — the latter a small Tokmok university, evidence that policy adoption is not confined to the capital's flagships), while the country's largest public universities (KNU, KSTU, Osh State, Manas) show no verifiable usage policy. Osh State does run an Institute of Nanotechnology and Artificial Intelligence, and Kyrgyzstan's government published a draft AI ethics regulation in September 2026 (dig.watch) — national-level momentum that may convert into institutional policies within a year.

Regional vs global comparison

Using the verified numbers from the companion Global University AI Policy Audit (187 universities, 14 October 2026):

Three things Central Asia shares with the rest of the world: (a) prohibition is dead — no campus-wide AI bans anywhere in the sample, matching the zero-ban finding across the global top 100; (b) the converging grammar of delegated permission and mandatory disclosure holds even in the region's small-university policies (IUCA's prompt-submission rule echoes practices seen from Utrecht to Tokyo); (c) adoption is fast-moving — four of the six policies date to 2024–2025, and national frameworks (Kazakhstan's 2024 standard, Uzbekistan's AI strategy to 2030, Kyrgyzstan's 2026 draft ethics regulation, Tajikistan's 2040 strategy) are all younger than two years.

What differs: the verified adoption rate is less than half of even the Global South flagship rate, and the UNVERIFIABLE share (81%) dwarfs the global audit's (~10%). Part of the difference is real (fewer published policies); part is methodological and structural (multilingual publishing, portal-gated documents, closed national webs). Both should be read as findings.

Limitations

1. The sample is 31, not the 30–50 target's upper end. Thirty-one meets the floor and is the largest verifiable Central Asian sample this program has assembled, but per-country subsets (3–10 institutions) are too small for precise national estimates. Treat country-level percentages as indicative.
2. UNVERIFIABLE is not absence. Twenty-five universities are marked UNVERIFIABLE, meaning a policy could not be located on the public official domain — including cases where a document exists but sits behind a portal (KBTU's ethics PDF could not be retrieved for content verification and was not counted), is published in a language the auditor could not read, or circulates only as scanned Russian/Kazakh/Uzbek PDFs that search engines do not index.
3. Turkmenistan is under-observed by design. The country's universities maintain minimal public web presences; the three institutions sampled are the most internationally visible ones. Findings for Turkmenistan describe verifiability, not institutional reality.
4. Two YES cases carry scope caveats (stated in the table): Toraighyrov's regulation covers PhD dissertations only; WIUT's is an institutional strategy with substantive teaching provisions rather than a standalone usage regulation. Excluding both would put the YES rate at 13% (4/31).
5. AUCA's policy page currently resolves from an archived subdomain (orient2020.auca.kg) with a reference to the live code-of-conduct page; its canonical live status could not be confirmed on 17 October 2026.
6. Secondary literature was excluded where unverifiable. Press reports (TimesCA on IITU's AI-tolerant thesis defenses; Al-Farabi press coverage) were used as search leads, not evidence. The one exception is national-context background, which is clearly labeled as such and never counted toward YES.
7. Snapshot date: 17 October 2026. Kazakhstan's AI-Sana program, Uzbekistan's 15-university AI-lab plan, and Kyrgyzstan's draft AI ethics regulation (public comment closed 13 October 2026) all suggest the 2027 picture will differ materially.

Sources

All URLs accessed 17 October 2026 unless noted. University-domain sources are listed in Tables 1–2. Secondary sources below are context only and do not support any YES classification.