China's Top AI Firms Restrict Tools During Gaokao

Leading Chinese artificial intelligence companies are implementing strict restrictions on their generative AI tools during the annual Gaokao (National College Entrance Examination). This coordinated move aims to prevent academic dishonesty by disabling image recognition and question-answering capabilities for exam-related content.

The policy affects millions of students and highlights the growing tension between educational technology and assessment integrity. Major players including ByteDance, Tencent, Baidu, and iFlytek have confirmed varying levels of service limitations during this critical period.

Key Facts About Gaokao AI Restrictions

• ByteDance's Doubao disables photo-based question answering but allows general usage during the exam window.
• Tencent's Yuanbao completely suspends services related to the Gaokao, including image recognition.
• Baidu's Wenxin Yiyan has not yet received official mandates but follows industry norms for fairness.
• iFlytek confirms that all major large language models will likely face operational constraints.
• Historical data shows similar restrictions were enforced in 2025 by Alibaba and Kimi.
• The primary goal is maintaining exam fairness and preventing real-time cheating assistance.

Industry Responses Vary by Platform

Customer support representatives from ByteDance clarified the specific boundaries of their tool, Doubao. They stated that while the platform remains accessible for general queries, features designed for solving exam problems via image upload are disabled. Users must rely on the interface display for real-time status updates, as no formal notification was issued beforehand. This approach balances accessibility with ethical responsibility.

In contrast, Tencent adopted a stricter stance through its Yuanbao assistant. A company representative confirmed that they explicitly prohibit answering questions during the examination period. Last year, they paused all service requests involving the Gaokao. This zero-tolerance policy ensures that no student can leverage the AI for an unfair advantage, reflecting a proactive compliance strategy.

Baidu and iFlytek Stances

Baidu, the developer of Wenxin Yiyan, reported that they have not yet received specific notifications regarding functional limitations for this year's exam. However, their business leaders acknowledge the importance of maintaining a level playing field. Their response suggests a wait-and-see approach, potentially aligning with regulatory directives if they emerge closer to the event date.

Meanwhile, iFlytek, a leader in speech and language processing, indicated that exact details remain pending. Nevertheless, their staff emphasized that any reputable large language model provider would naturally impose restrictions. This consensus among tech giants underscores a unified industry effort to uphold educational standards without waiting for external enforcement.

Historical Precedents and Technical Measures

Looking back at the 2025 Gaokao provides clear insight into current trends. Reports from IT Home indicate that multiple popular AI tools suspended image recognition functions. Alibaba's Tongyi Qianwen and ByteDance's Doubao stopped responding to试卷 (exam paper) images. This technical block prevents users from uploading photos of test questions to receive instant solutions.

Other platforms took even more drastic measures. Kimi and Tencent's Yuanbao completely halted image recognition services during the exam hours. When users attempted to use these features, they encountered messages stating that the function was unavailable to ensure fairness. These precedents suggest that total suspension is becoming the standard for high-stakes testing periods.

Comparison of Restriction Levels

The variation in responses highlights different risk management strategies among tech firms:

• Partial Restriction: Disabling only photo-to-answer features while keeping chat active.
• Total Suspension: Halting all exam-related interactions and image processing.
• Regulatory Alignment: Waiting for government directives before implementing changes.
• Proactive Ethics: Self-imposing bans to maintain public trust and brand integrity.

Broader Implications for EdTech and AI

This situation reflects a global challenge facing the education sector. As generative AI becomes more capable, traditional assessment methods face obsolescence. Western institutions also grapple with similar issues, often relying on plagiarism detectors or proctoring software. However, the Chinese approach involves direct intervention by tech providers at the source.

For developers, this sets a precedent for responsible AI deployment. It demonstrates that commercial interests may need to yield to societal values during critical events. Companies must build flexible systems that can rapidly adapt to such requirements without compromising overall service stability.

Furthermore, this impacts user behavior. Students and educators must understand that AI is not an unrestricted resource. The temporary blackout serves as a reminder of the ethical boundaries surrounding academic work. It encourages reliance on personal knowledge rather than algorithmic shortcuts during high-pressure scenarios.

What This Means for Stakeholders

Students should prepare for exams without assuming AI assistance will be available. Relying on these tools for last-minute study might lead to frustration if features are unexpectedly disabled. Educators can view this as a validation of human-centric assessment, at least for now.

For businesses, maintaining compliance is crucial. Failure to restrict access could result in reputational damage or legal repercussions. Proactive communication with users about these limitations helps manage expectations and reduces support ticket volumes during the exam period.

Developers must prioritize modular design in their AI applications. Being able to toggle specific features like image recognition quickly is essential. This agility allows companies to respond to regulatory changes or ethical concerns without rebuilding their entire infrastructure.

Looking Ahead: Future of AI in Education

The temporary ban on AI tools during the Gaokao is likely just the beginning. As AI technology advances, the definition of "cheating" may expand. We might see permanent restrictions on certain types of academic assistance or the development of specialized, monitored AI tutors for educational use.

Regulators worldwide will watch these developments closely. The success of these restrictions in China could influence policies in Europe and North America. If effective, it may lead to standardized protocols for AI usage during certified assessments globally.

Ultimately, the goal is to integrate AI into education responsibly. Rather than banning it entirely, the focus should shift toward enhancing learning outcomes while preserving the integrity of evaluations. This balance requires ongoing dialogue between technologists, educators, and policymakers.

Gogo's Take

• 🔥 Why This Matters: This marks a pivotal moment where Big Tech voluntarily cedes functionality to protect social institutions. It proves that AI companies recognize their role in societal infrastructure and are willing to prioritize fairness over engagement metrics during critical windows.
• ⚠️ Limitations & Risks: While well-intentioned, these bans are reactive. They do not solve the underlying issue of how to assess knowledge in an AI-saturated world. There is also a risk of creating a black market for unregulated, offshore AI tools that lack these ethical guardrails.
• 💡 Actionable Advice: Educators should immediately audit their assessment methods for AI vulnerability. Instead of fighting AI, integrate it into the curriculum with clear guidelines. For students, focus on developing critical thinking skills that AI cannot easily replicate, ensuring long-term relevance in the workforce.