In a rare but telling pause, Chinese tech heavyweights Alibaba and Tencent have temporarily nerfed some core features of their AI chatbots—right in the middle of one of the most stressful weeks in China: gaokao season.
If you’re not familiar, the gaokao (June 7–10) is China’s national college entrance exam, and it’s the definition of high stakes. We’re talking millions of students competing for a shot at the country’s top universities, often with lifelong consequences tied to that score. No pressure, right?
To help preserve the integrity of this make-or-break exam, Alibaba’s Qwen and Tencent’s Yuanbao have both disabled features like image recognition. That’s a big deal. These tools are normally used for everything from translating menus to helping decode math problems. But during exam week, they could be exploited for some shady, photo-based cheating.
Other AI players like Moonshot’s Kimi joined the effort by pulling the plug on similar capabilities, while ByteDance’s Doubao kept its image features online but slapped on stricter filters. The message from China’s tech sector? We’re not letting AI turn into an exam loophole.
This is more than just a PR move. It shows how seriously China is taking AI governance. Regulators there are threading a needle—encouraging innovation, but with hard limits when ethics and fairness are at risk. Education officials are even introducing AI literacy into school curricula by 2025. But they’ve also made it clear: no AI-generated homework, and definitely no chatbot-powered test answers.
The gaokao isn’t just a test. It’s the cornerstone of China’s meritocratic system. For rural or low-income students, it’s often the only way up. So any hint of AI-assisted cheating could do real damage, not just to individual outcomes but to the social contract that holds the whole system together.
What’s wild is how aligned this response has been—government policy and private tech are in sync. Instead of waiting for laws to catch up, companies are taking real-time action. That includes Alibaba, which—even while pausing certain Qwen features—continues to push the envelope with its open-source AI models like Qwen3 Embedding, now fluent in over 100 languages.
Bottom line: this isn’t a step back for AI. It’s a lesson in responsible deployment—because knowing when to pause is just as important as knowing how to build.
