China's AI models censor history, filter criticism and push propaganda, report finds

By Robert Stanley | 2025-03-25

China is embedding its authoritarian values into artificial intelligence (AI) systems, using them as a tool to reshape "authoritarian narratives," undermine global democratic norms and dominate critical global industries, according to a recent report.

China's first-mover advantage in AI, coupled with its open-source tools that are outperforming Western models on key benchmarks and at dramatically lower costs, has accelerated their global adoption, the American Edge Project, a pro-business policy group, wrote in a December 2024 report.

This situation is "threatening US leadership and global security," it noted.

Chinese AI models, which the report states are specifically designed to conform to authoritarian narratives, could become enablers of misinformation and oppression.

'It's manipulation'

These systems censor historical events, justify human rights abuses and promote state propaganda, according to the paper.

"AI systems that conform to authoritarian narratives represent a grave threat to global freedom and democracy," the report's authors warned. "This is not innovation -- it's manipulation."

Their solution: "America must lead in setting the global standard for objective, truth-driven AI to ensure that this transformative technology reflects democratic values and strengthens free societies," the report said.

The report compares several leading Chinese AI models with their American counterparts, illustrating how the two sides differ in handling sensitive topics.

For example, when asked about the Tiananmen Square massacre of 1989, Chinese models like Hunyuan-Large, Qwen2-72B and ChatGLM-4 either refused to answer or downplayed the deaths that occurred.

Filtering criticism

In contrast, US models, such as Grok 2 + Flux, Meta's Llama and ChatGPT, provided detailed, factual accounts of the crackdown by Chinese forces, including historical and human rights context, the report said.

The same pattern emerged in questions about human rights, with Chinese models either censoring or muting criticism of China's record while offering full responses to critiques of the United States.

The Chinese models even filtered criticism of China's political leaders, conceding "I can't comply with that request," when asked to criticize Chinese President Xi Jinping, according to the report.

These discrepancies highlight the stark contrast in how Chinese and US AI models handle politically sensitive information, with China's systems intentionally obscuring or distorting historical facts to align with the Communist Party's narrative, it said.

This strategy, the report argues, is part of Beijing's broader push to use AI as a tool to reshape global discourse in its favor.

China is using its affordable and increasingly ubiquitous AI models as a "Trojan horse" to sneak Chinese values into the global information system, the report adds.

Surveillance, censorship and misinformation

"A world of unchecked, Beijing-built AI ecosystems would be a major blow to the US and to humanity writ large," the report noted, citing a 2023 assessment by the Center for a New American Security.

"If Chinese AI goes global, so too will brazen noncompliance with international agreements on the technology," it said, adding that China's open-source tools are "directly wiring the CCP [Chinese Communist Party]'s values of surveillance, censorship, misinformation and control into the world's technological foundation."

As US and European policymakers grapple with how to regulate AI, the report stresses the need for the United States to lead in setting global standards for truth-driven AI that reflects democratic values.

To maintain its competitive edge, the United States must secure long-term investments in key areas like education, advanced chip production and cybersecurity, while fostering public-private partnerships to ensure that American AI remains the preferred choice globally, the paper argues.