China has embraced AI to help it censor and surveil its citizens and is exporting its techniques to the world, according to a new report by think tank the Australian Strategic Policy Institute (ASPI).

Titled “The party’s AI: How China’s new AI systems are reshaping human rights” and published yesterday, the report observes that in Europe and the US the concept of “AI safety” is largely understood as making the technology safe and fair.

In China, the report says, the definition of AI safety is “ensuring that AI serves ‘core socialist values’ and the political stability of the state.”

The report says AI helps china to meet those goals in several ways, one of which is censoring large language models so they refuse to respond, omit sensitive details, or restate official narratives. ASPI tested four Chinese AIs – Baidu’s Ernie Bot, Alibaba’s Qwen, Zhipu AI’s GLM and DeepSeek’s VL2 – using a dataset of images depicting the 2019 Hong Kong protests, the Tiananmen Square protests and related memorials, leaders of the Chinese Communist Party, Falun Gong demonstrations, and eight other sensitive topics.

“Those tests show that Chinese-developed models display stronger censorship behaviours in response to politically sensitive imagery than their US-developed counterparts,” the report found. “The most direct censorship behaviour was an outright refusal to respond, which was especially common in models accessed using inference providers headquartered in Singapore rather than the US, where sensitive prompts frequently triggered error messages or blank outputs.”

ASPI thinks those non-responses matter because Chinese AI is becoming accessible and popular around the world.

“The threat lies less in overt propaganda than in quiet erasure, when the machine that describes reality begins deciding which parts of reality may be seen,” the report states.

Erasure is already happening online, because China has made it plain that its publishers and web giants should use AI to filter material on their platforms.

“In China, AI now performs much of the work of online censorship, scanning vast volumes of digital content, flagging potential violations and deleting banned material within seconds,” the report finds. China requires its web giants like Tencent, Baidu and ByteDance to develop AI to filter content, and those companies have productized their efforts and therefore become part of the AI-powered censorship apparatus.

The report notes that China’s censorship regime still needs human content reviewers, because AI can’t yet interpret satire, keep up with evolving idioms, or understand all minority languages.

China has therefore “effectively ‘deputized’ small and medium-sized enterprises, under the principle of ‘self-discipline’ (自律), to police their users on behalf of the authorities, thus showing how they adopt AI tools and train censorship workers in a system in which human judgement remains indispensable – for now.”

• China cut itself off from the global internet for an hour on Wednesday
• China tells grumps, trolls, and AIs to stop emoting online
• Chinese censorship-busters claim Tencent is trying to kill its WeChat archive
• China could be doing better at censorship, think tank finds

ASPI also feels that AI has become pervasive in China’s justice system, and made it less accessible and fair.

“A criminal suspect in China may be identified and arrested with the aid of the world’s largest AI-powered surveillance network; be prosecuted in courts that use AI to draft indictments and recommend sentences; and, finally, be incarcerated in a prison facility where AI-enabled surveillance systems extensively monitor their emotions, facial expressions and movements, feeding data back to central monitoring,” the report states.

All that AI, the report says, “enhances the efficiency of police, prosecutors and prison administrators, reduces transparency and accountability, and further enables state repression.”

ASPI’s authors note that China already practices “collective punishment and cultural erasure of Uyghurs and Tibetans.” AI trained on those practices will replicate those policies. China is also ensuring its AIs understand minority languages like Uyghur, Tibetan, Mongolian and Korean so that its AI-powered surveillance and censorship tools are more effective.

Stealing food

ASPI also suggests Chinese AI is impacting people well beyond the nation’s borders by making it harder for them to find food.

“The Chinese Government is actively supporting a national fishing industry that routinely violates the economic rights of the citizens of at least 80 countries,” the report states,” before accusing China of overfishing some waters “into a state of severe depletion, flouting reporting requirements and harassing local fishers.”

And now China is doing that with AI.

“Chinese fishing fleets are now adopting AI-powered intelligent fishing platforms that further tip the technological scales towards Chinese vessels and away from local fishers,” the report states. “That use of AI amplifies China’s state-supported erosion of the economic rights of affected communities, to the financial benefit of Chinese private and state-owned companies – even as China continues to tout its support for the economic rights of developing countries.”

The report details three AI-powered fishing platforms, one called “’AoXin 1.0’ that is a deep-learning-based prediction model that provides near real-time forecasts of squid fishing grounds.”

AoXin apparently helped to identify four new squid fishing grounds, and improved catch rates on ships that use it by 15 to 20 percent.

“The system is deployed on board with AI hardware, using the Huawei Ascend AI chip,” ASPI found.

The think tank believes China has found “particularly willing buyers for its AI-enabled technology in autocracies and weak democracies" among developing nations of the “global South.”

Beijing is also trying to make its approach to AI a global norm by participating in governance and standards bodies.

ASPI suggests other nations need work to “prevent China’s AI models, governance norms and industrial policies from shaping global technology ecosystems and entrenching digital authoritarianism.” ®