Apple Intelligence must pass China's AI censorship test

This article may contain affiliate links. Learn more on our Partner Program page.

Make Apfelpatient a preferred source One click – and you'll see us more often on Google

Apple Intelligence is subject to special legal and political regulations in China. Unlike in Western markets, the use of AI there is determined not only by its technical capabilities, but above all by whether it fully complies with government regulations on information control. For Apple, this means an extensive review process that delves deeply into the workings of Apple Intelligence and sets clear limits on what content may be accessible.

China has been rigorously controlling the flow of digital information for years. This control no longer extends solely to traditional search engines or social networks, but increasingly also to AI systems. Because AI models can aggregate, explain, and classify information, they are considered particularly sensitive by the authorities. Apple Intelligence is therefore rigorously tested and regulated before its release in the Chinese market.

Chinese censorship as a starting point

The Chinese government maintains control over online information through extensive filtering mechanisms. Many Western platforms are inaccessible in the country. Google withdrew from China after being forced to censor search results. Services like Facebook, X, and Wikipedia are also blocked. Even when using the Chinese search engine Baidu, numerous search terms are filtered or not displayed at all.

AI systems could theoretically circumvent these restrictions by drawing on knowledge from training data or indirect sources. This is precisely why they are also subject to strict monitoring. Foreign companies are not allowed to freely deploy their own models in China but must use approved Chinese AI models.

Apple Intelligence and China's Special Role

Internationally, Apple is collaborating with OpenAI on Apple Intelligence. If Siri cannot answer a request itself, it is forwarded to ChatGPT. Additionally, it has been revealed that Apple intends to use customized Google Gemini models from Google on its own Apple servers to process Siri's responses in the future.

This approach is not permitted in China. There, Apple was effectively forced to enter into a contract with a Chinese AI provider. At the beginning of the year, it was revealed that Apple had partnered with Alibaba. Alibaba's AI model, Qwen3, a competitor to ChatGPT, was specifically optimized for use on Apple devices and is subject to Chinese censorship regulations.

The mandatory test with 2,000 questions

As the Wall Street Journal reports, all AI models operated in China must pass a government-mandated test before being publicly released. This test is designed to ensure that the systems do not grant access to prohibited or politically sensitive information.

The models are confronted with 2,000 targeted questions. These questions aim to retrieve content that is censored or considered problematic within China. The chatbot must reject at least 95 percent of these queries. It is not allowed to provide answers that could undermine state authority, question official government propaganda, or be classified as discriminatory.

The regulations also require that these sets of questions be updated at least once a month. This means that the models must be regularly adapted and retested.

A whole industry dedicated to AI censorship

Preparing for this test is so demanding that an entire industry has sprung up in China to focus on it. Specialized agencies help AI companies train and configure their models to meet government requirements. Insiders compare this process to preparing for a challenging university entrance exam, where students learn precisely which answers to avoid.

The inherent contradiction of the specifications

The task is further complicated by the fact that the Chinese government is pursuing two conflicting goals. On the one hand, information is strictly censored domestically, meaning AI models are not allowed access to much content. On the other hand, these models are expected to be powerful and internationally competitive. This would actually require broad access to information, including content from websites that are blocked within China.

The responsibility for technically implementing these conflicting requirements is left to the AI companies themselves. They must ensure that prohibited information is reliably filtered without unduly limiting the performance of the models.

Political regulations place strict limits on Apple Intelligence

Apple Intelligence's presence in China clearly demonstrates the significant impact political regulations can have on modern AI systems. To be allowed to manufacture and sell products in the country, Apple makes compromises that are difficult to reconcile with its own principles in other markets. Issues such as human rights violations by the Chinese government are particularly high on the list of topics Apple Intelligence is prohibited from addressing in China. This case makes it clear that technological innovation alone is insufficient when state control dictates the framework. (Image: Shutterstock / songweiqiang)