Recently, several major large model platforms in China responded to the online rumor that "AI tools will be disabled during the Gaokao period," stating that some functions of multiple AI tools will be strictly restricted during the upcoming Gaokao to maintain exam fairness.

According to reports, Douyin's Duanbao customer service confirmed that it can be used normally during the Gaokao period, but functions such as taking pictures to answer questions will be disabled; an internal person from Tencent revealed that its AI assistant Yuanbao has already clearly adopted a strategy of not answering questions during last year's Gaokao. Historical data also shows that Yuanbao restricted users' image recognition and summary instructions on June 9th last year through a unified response.

Learning, Exams, Gaokao, Education (1)

At the same time, while the relevant personnel from Baidu Wenyi Yiyuan and iFLYTEK stated that they have not yet received specific notifications about function restrictions, iFLYTEK technical staff emphasized that, according to industry consensus, large models generally take restriction measures during the Gaokao period. Currently, core AI tools including Kimi, DeepSeek, and Duanbao have completed relevant compliance settings for the Gaokao period.

This move reflects the collective self-discipline and shift toward compliance in the generative AI industry before major social exams. With the popularity of multimodal image recognition and strong reasoning models among students, the boundary between "learning assistance" and "exam assistance" provided by AI tools has become a key focus for regulation and engineering protection. The active or passive disconnection of question-answering functions by large model platforms during the Gaokao is not only a compliance response to prevent cheating, but also indicates that AI applications are rapidly entering a mature stage of precise governance by scenario and time period.