Reported Student Misuse of ChatGPT, StudyX, and Gemini to Obtain Answers During Vietnam's 2025 National High School Graduation Exam

Multiple students in Vietnam reportedly used generative AI tools including ChatGPT, StudyX, and Gemini to cheat during the national high school graduation exams on June 26–27, 2025. Incidents reportedly included smuggling phones and cameras into exam rooms, transmitting questions outside in real time, and receiving purportedly AI-generated answers. Students were caught at multiple centers in Lam Dong and Hanoi. Authorities are investigating the breaches and related information leaks.

Risk classification

• Primary risk domain: 4 Malicious actors
• Primary risk subdomain: 4.3 Fraud, scams, and targeted manipulation

The students used AI applications to cheat on high-stakes national exams, which constitutes using AI to gain an unfair personal advantage through cheating.

Causal factors

• Entity: Human
• Intent: Intentional
• Timing: Post-deployment

The cheating was initiated by the intentional actions of human students using deployed AI systems to gain an unfair advantage.

EU AI Act risk tier

• Risk tier: 3 Limited Risk

Limited Risk: The AI systems used (ChatGPT, Gemini) are general-purpose chatbots, which fall under the Limited Risk category requiring transparency.

AI system and alleged parties

• AI system: ChatGPT, Gemini, StudyX (Google, OpenAI)
• AI purpose: Question Answering; Writing Assistant
• Behaviour type: Assistant
• Alleged developer: StudyX, OpenAI, Google
• Alleged deployer: Unnamed students in Vietnam accused of cheating
• Alleged harmed parties: Vietnam Ministry of Education and Training, Universities relying on exam scores, Students in Vietnam, General public of Vietnam

Harm severity

Severity ratings are based on this scale.

Highest direct severity in any category: Negligible. Severity is scored from Negligible to Catastrophic in each harm category, for harm the reports describe as caused directly or indirectly by the AI system.

• Physical: direct Negligible, indirect Negligible
• Infrastructure: direct Negligible, indirect Negligible
• Property: direct Negligible, indirect Negligible
• Financial: direct Negligible, indirect Negligible
• Environmental: direct Negligible, indirect Negligible
• Malicious content: direct Negligible, indirect Negligible
• Differential treatment: direct Negligible, indirect Negligible
• Civil rights: direct Negligible, indirect Negligible
• Democracy: direct Negligible, indirect Minor
• Privacy: direct Negligible, indirect Negligible
• Psychological: direct Negligible, indirect Negligible
• Epistemic: direct Negligible, indirect Negligible
• Child sexual exploitation and abuse: direct Negligible, indirect Negligible

Democracy

Reported: The reports describe concerns about the security of examination materials and the integrity of the national exam process, which is a key public institution.

Indirectly caused: The integrity of the national high school graduation exam, which determines university admissions for 1.16 million students, was compromised by the leak of exam questions to AI apps.

People affected

• Occurrences reported: 6
• People reportedly harmed: 6
• People reportedly exposed: 1160000

Potential causes

Management

• Insufficient Exam Security Planning: Organizers failed to anticipate high-tech AI cheating methods.
• Inadequate Proctor Training: Staff were not trained to identify modern electronic cheating devices.
• Lack of Incident Response Protocol: A student was caught but not suspended immediately, allowing repeat offenses.

Technology

• AI-Powered Question Solving Apps: Apps like ChatGPT and Gemini allow quick generation of exam answers.
• Miniaturized Streaming Hardware: Button-sized cameras and tiny earphones enable covert data transmission.
• Accessible Mobile AI Interfaces: Easy-to-use mobile apps make real-time cheating highly accessible.

Data Inputs

• Captured Exam Paper Images: High-resolution photos of test papers provided clear input for AI solvers.
• Text-Based Exam Questions: Standardized questions are easily parsed and solved by large language models.
• Real-Time Video Streams: Live streams of exam sheets allowed external helpers to feed data into AI.

Human Factors

• Academic Pressure on Students: High-stakes university entrance exams drive students to use illicit aids.
• Negligent Exam Proctors: Proctors failed to detect smuggled phones and high-tech streaming gear.
• Collaborator Assistance: External accomplices facilitated the input of questions into AI systems.

Process and Methods

• Weak Candidate Screening: Lack of physical searches allowed students to smuggle devices into rooms.
• Inadequate Proctor Supervision: Proctors failed to monitor students actively, allowing photo taking.
• No Signal Jamming in Exam Centers: Absence of signal blockers enabled real-time internet communication.

Regulatory Environment

• Unregulated Online Device Sales: Spy cameras and earpieces are easily purchased on social media platforms.
• Lack of AI Abuse Frameworks: No specific regulations exist targeting AI-assisted academic dishonesty.

Information quality

• Classification confidence: High
• Reason for confidence: The reports clearly detail the cheating incidents, the specific AI tools used (ChatGPT, StudyX, Gemini), the methods of transmission, and the consequences for the students. There is little ambiguity regarding the facts of the cheating.

During Vietnam's 2025 national exams, students used AI tools like ChatGPT and Gemini to cheat, prompting police investigations into the leak of classified exam materials. While the incident disrupted a major government-run educational process and caused public concern, its national security impact remains minor and manageable within standard law enforcement and administrative procedures.

• Overall national security impact: Minor
• Scope: Single nation
• Primary target: Vietnam
• Alleged perpetrator: Individual students

Threat characteristics

• Imminence: Long-term. Represents an ongoing strategic concern regarding exam security and academic integrity rather than an active crisis.
• Autonomy: Human-controlled. AI tools acted strictly as assistants under direct human instruction to solve specific questions.
• Novelty: Evolved capability. Represents an evolution of traditional cheating methods through the integration of real-time AI vision and text generation technologies.

Impact by dimension

• Physical security: Negligible. No threat to physical systems, critical infrastructure, or human safety identified in the incident.
• Information security: Negligible. No compromise of national intelligence capabilities, military secrets, or systematic information warfare operations.
• Sovereignty: Minor. Disruption of a major state-administered national exam and leak of classified exam papers, prompting a police investigation, but manageable within standard procedures.
• Economic security: Negligible. No threat to financial systems, strategic industries, or technological competitive advantage identified.
• Societal stability: Minor. Public concern regarding the integrity of a high-stakes national exam affecting 1.16 million students, though it did not threaten overall social cohesion.