When our team encountered a microcamera built into an examination keyboard, it was tempting to frame the problem as a contest between cheating devices and detection tools. Our experience in Kazakhstan showed that the larger challenge was the path from a candidate's admission to an accepted result. A secure camera feed could not repair an admission process that allowed assistance. A verified identity could not establish that an answer was independently produced.
We developed TrustExam.ai, deployed locally as PDD 3.0, and initiated changes to Kazakhstan's driver theory examination workflow. Kazakhstan is a Central Asian country between Russia and China. The public operator, Government for Citizens, administers testing centers; the knowledge test discussed here is separate from practical driving. Five lessons from the implementation may be useful to other assessment providers.
1. Map opportunities for influence before choosing controls
Before the national rollout, an August 2022 police investigation in Aktobe, western Kazakhstan, described staff admitting selected candidates with prohibited equipment while accomplices supplied answers. The alleged fee was KZT 200,000, approximately US$420 at the historical exchange rate. [Police account](polisia.kz/ru/opg-zanimalas-nezakonnoj-vydachej-voditel-skih-udostoverenij-v-aktobe).
That is a process problem as well as a device problem. A control loses value when an individual can decide who must comply with it. We examined opportunities for assistance both at testing centers and in central administration, including admission and assignment to a particular workstation.
Our response combined a decentralized architecture with changes to admission, monitoring and result handling. We sought to minimize dependence on individual employees throughout the workflow. For other providers, the useful exercise is to trace who can influence each step and what record that influence leaves.
2. Connect identity to an examination attempt
In the process we were rebuilding, earlier biometric announcements had not become an effective operational control. Our implementation experience included the possibility of a passing result without the candidate entering the examination room.
We made biometric verification part of the actual workflow, connecting the candidate's identity, application, assigned workstation and session. Audio and video monitoring, automated event analysis and answer records then supplied material for a separate review after a passing score.
The design question is whether a reviewer can follow the evidence from the person admitted to the session completed and the result accepted. A successful identity check is one part of that chain.
3. Turn signals into reviewable evidence
A case reported in November 2023 illustrates the difference between an unusual event and an actionable explanation. In Shymkent, southern Kazakhstan, an accomplice tapped on a window. The operator said monitoring captured repeated turns and movements toward it, while the audio recording captured the tapping. The result was annulled. [Operator account reported by Informburo](informburo.kz/novosti/podskazyval-otvety-postukivaya-po-steklu-na-kakie-uxishhreniya-idut-kazaxstancy-ctoby-sdat-na-prava).
Our workflow allowed authorized staff to consider those observations together. AI proctoring could identify events for attention, while reviewers considered context and applied the operator's rules. An alert alone did not establish misconduct.
This distinction also matters in analysis across centers. Keep an automated signal, a completed review and a confirmed violation as separate events. Use consistent reporting periods and distinguish examination attempts from people. These are principles for interpreting the evidence, not a claim that an unusual statistical pattern proves cheating.
4. Expect the threat to change after launch
The common attempts we encountered involved miniature cameras and micro earpieces. We also encountered vibration and tapping used to communicate answers. The people organizing assistance had a financial incentive to keep testing the controls.
We introduced micro earpiece detection; the operator described the electromagnetic detection equipment publicly in February 2024. More recently, Bluetooth and Wi-Fi scanning added information about nearby wireless devices. Proximity can help direct an inspection, but does not establish ownership or use. Detector coverage varies. [Operator announcement](gov4c.kz/ru/press-center/news/343532).
Record integrity needed similar attention. We strengthened application protection, cryptographic integrity checks and controls on changes to stored examination records. We also continued reliability work, including workstation readiness checks and recovery from temporary connection problems. An interruption must be distinguishable from misconduct.
5. Report outcomes within their actual scope
The October 2023 national launch covered 83 examination rooms. The operator reported 66 results annulled during the first month and 222 dishonest-test-taking cases identified cumulatively by February 20, 2024. The latter includes the earlier period. [Launch report](informburo.kz/novosti/podskazyval-otvety-postukivaya-po-steklu-na-kakie-uxishhreniya-idut-kazaxstancy-ctoby-sdat-na-prava), [February figures](gov4c.kz/ru/press-center/news/343532).
Those are dated operational outcomes. They are not a current fraud rate, a before-and-after estimate of corruption or evidence about the effectiveness of the recent wireless layer. The operator's later risk analysis identified continuing weaknesses across several systems, including PDD 3.0. Launch began an ongoing improvement process. [Risk analysis](gov4c.kz/upload/iblock/76d/ax9ccmsq8ugaeg615abdt2t0kap0v5gu.pdf).
Applying the lessons to the next format
As of October 2026, we are preparing online driver theory examinations and a planned transition to AI-led decisions during and after the session. Our public-sector counterparts' experience of the technology and pressure on physical testing capacity inform that work. The existing center-based outcomes came from a workflow with staff review; they do not validate a completed online or autonomous rollout.
For assessment providers considering similar changes, start with one examination journey. Identify every point where a person can influence admission, the session or acceptance of the result. Then ask what evidence would allow a decision to be reviewed. That exercise makes a procurement discussion much more concrete than a list of detection features.
Author disclosure: Yersultan Yermanov is the founder of TrustExam.ai, the technology provider described here. This article draws on our [published case study and linked sources](trustexam.ai/blog/kazakhstan-driver-license-knowledge-test-integrity-case-study). It is an account from the implementation team.
Currency note: the approximate US dollar conversion uses KZT 476.43 per US$1 on August 22, 2022. [National Bank of Kazakhstan rate]
Join the global community advancing e-assessment through innovation, research, and collaboration.
Keep informed