Chinese tech giants, including ByteDance, Tencent, Alibaba, and Moonshot, have temporarily blocked access to their generative AI services during the national entrance exams known as Gaokao. These measures aim to prevent the use of artificial intelligence for cheating and other forms of academic dishonesty among over 13 million students taking the exams from June 7 to 10, 2025. Gaokao results are crucial for university admissions in China.
This is reported by Business • Media
AI Services Temporarily Unavailable for Students
Users report that popular AI-based services — Doubao from ByteDance, DeepSeek, Yuanbao from Tencent, Qwen from Alibaba, and Kimi from Moonshot — have disabled question-answering and image recognition features. Some applications displayed messages about temporary unavailability due to the exams. Despite attempts to bypass filters, AI programs refused to respond even to modified queries. This has sparked outrage among students who regularly use such tools for studying. Some jokingly blamed the Gaokao candidates for blocking access to the services.
Increased Security Measures During Exams
In addition to restricting AI services, Chinese authorities have significantly enhanced both digital and physical security during the exams. In various regions, AI tools have been implemented to monitor suspicious activities: special cameras track student behavior, and the collected data is analyzed after the exams are completed. Biometric identification, digital device checks, and radio signal jamming using specialized “jammers” are employed for additional control. Access rules to exam venues have also been tightened.
In some cities, transportation has been further restricted, and mass events have been postponed to ensure students arrive at exam locations on time. All these measures are part of the government’s strategy to ensure transparency and fairness in Gaokao.
“AI services from ByteDance, Tencent, and Alibaba are suspended during the exams.”
Previously, Anthropic’s CEO Dario Amodei expressed serious concerns about data security at the Chinese startup DeepSeek. According to him, their AI model R1 showed the worst results in information protection tests.