A camera concealed above the F8 key of an examination keyboard could send a test question to someone outside the room. The candidate could appear to be working independently while receiving help. Our team encountered this device while securing driver theory examinations in Kazakhstan.
Microcamera concealed inside an examination keyboard. Photo provided by the TrustExam.ai team.
The challenge extended beyond spotting prohibited equipment. We needed to connect the candidate’s identity, examination session and supporting evidence to the decision about the result. We also had to reduce opportunities for employees to influence that process.
TrustExam.ai, deployed locally as PDD 3.0, brought these steps together. Our team initiated the process changes and built the platform; the public operator ran the examinations and made the official decisions. The system supports a national programme whose analysed operational records contain more than one million exams with a recorded pass/fail outcome.
The programme at a glance
Setting: Driver licensing examination centres in Kazakhstan, operated by Government for Citizens.
Scope: Computer-based theory examinations. Practical driving assessments are separate.
Approach: Biometric identity checks, examination monitoring, staff-access controls and review of session evidence.
Scale in the analysed records: 1,073,717 exams with a recorded outcome, covering 2023 through 12 August 2026.
More than one million exams, with a defined counting method
Our analysis of the available operational database counted non-test, non-deleted examination records with a recorded pass/fail outcome. It counts exams, including repeat sittings, rather than unique people or the separate technical records created during a session.
Period | Exams with a recorded outcome |
|---|---|
2023 | 50,628 |
2024 | 336,568 |
2025 | 339,837 |
2026, through 12 August | 346,684 |
Total in the analysed records | 1,073,717 |
Source: TrustExam’s analysis of operational records. The latest non-test record in this dataset is dated 12 August 2026. These are historical volumes, not a live October total or an independent audit of performance.
The 2026 count had already exceeded the full-year 2025 count by that cutoff. This illustrates the volume the programme was handling. It does not establish how many violations were prevented, or prove that demand increased because of TrustExam.
Changing how a passing result becomes official
Our implementation experience exposed a basic weakness: in the process we were rebuilding, a passing result could be obtained without the candidate entering the examination room. Biometric verification had been announced earlier, but it was not an effective operational control in that workflow.
We redesigned the process to reduce its dependence on individual employees, from local centres to central administration. A decentralized architecture, controlled access and records of staff actions supported that change. Assigning a candidate to a particular computer or allowing prohibited equipment into the room could otherwise compromise an exam before its first question.
Admission. Link the candidate’s biometric identity check and application to the assigned workstation and session.
Examination. Collect audio, video and detected events so the session can be examined in context.
Review. Give authorized staff access to answers, recordings and relevant events when assessing a passing result.
Decision. The operator’s authorized staff apply the programme’s rules and accept or reject the result.
An AI alert directs attention to something worth checking. It does not establish misconduct by itself. Connecting these steps gives reviewers a basis for deciding whether a passing score should stand.
PDD 3.0 (TrustExam.ai): the exam review interface brings together session recordings, question-level results and an event log. The camera preview is redacted; event flags are prompts for review, not confirmed violations. View full-size image.
Detecting assistance that video alone can miss
The attempts we encountered included hidden earpieces, miniature cameras, improvised vibrating devices and accomplices communicating through taps. A clear view of a candidate’s face could still miss the equipment supplying answers.
We introduced micro-earpiece detection alongside the examination controls. In February 2024, the operator described detectors for low-frequency electromagnetic signals. More recently, we added Bluetooth and Wi-Fi scanning to help identify nearby devices that may warrant inspection.
These methods cover different technologies. Bluetooth scanning cannot detect every kind of earpiece, and a nearby signal does not establish who owns or uses a device. Staff need to consider the reading alongside the session evidence and, where appropriate, inspect the equipment. The newer wireless layer was not responsible for results reported at the initial launch.
We also encountered attempts to interfere with communications and examination records. Application protection, cryptographic integrity checks and restrictions on changes to stored records help protect the evidence used in review. Staff activity logging and two-factor authentication for operational roles support accountability within the service.
What the early results demonstrate
The operator announced the Akmola-region pilot in September 2023. At national launch in October, the system covered 83 examination rooms. The operator reported 66 annulled results in the first month and 222 identified cases cumulatively by 20 February 2024. These are overlapping historical totals, not figures to add together. First-month report; February update.
One reported case in Shymkent illustrates the review process. An accomplice tapped on a window to communicate answers. The operator described the candidate’s repeated turns toward the window and tapping captured in the audio. The result was annulled. Several observations could be considered together even though there was no visible phone or spoken answer.
These outcomes show that examination evidence could support action on individual results. They do not establish a current fraud rate or a percentage reduction in corruption. Our programme-volume figures and the operator’s early violation figures describe different periods and should not be combined into such a rate.
Maintaining the controls after rollout
People selling assistance had an incentive to keep finding weaknesses. We continued developing the system as their methods changed. The operator’s analysis of 2024–2025 risks identified remaining weaknesses across several systems, including PDD 3.0. National deployment did not mean that every vulnerability had been eliminated.
Reliability was part of that work too. Workstation readiness checks and recovery from temporary connection problems help separate technical interruptions from suspicious conduct. Security controls need to preserve a fair examination experience as well as produce useful evidence.
The next stage: preparing for online theory exams
As of October 2026, we are adapting the platform and workflow for online driver theory testing. The intended benefit is wider access beyond the capacity of physical examination rooms. In our work with the public operator, we have seen readiness to consider a greater role for technology based on experience with the existing deployment.
The planned model would give AI a decision-making role during the exam and in determining the outcome afterward. It remains a development direction. The operating model described in this case combines automated signals with staff review, and the historical results above come from that deployment. Practical driving assessment remains separate.
What other examination authorities can apply
Start by mapping how an exam result becomes official. Check whether the candidate is reliably linked to the session, whether a reviewer can reconstruct what happened, and where an employee could influence the process without a reviewable record.
A useful pilot should test those questions with both ordinary sessions and disputed events. It should also show what happens when equipment or connectivity fails. The value lies in evidence that supports a defensible decision and controls that staff can use consistently across examination centres.
If you operate a licensing or certification programme, explore TrustExam’s driver theory exam solution or book a demonstration to discuss the workflow and evidence your pilot should verify.