US and China battle to control how generative AI shapes our truth

America and China are locked in a high-stakes race to control the very systems that will teach billions how to see the past, understand politics, and decide who to trust. Once libraries handed out ten books or Google offered ten links, users got choices. Now generative AI delivers one answer before we even ask the question. It reads everything, synthesizes facts, and feeds us a synthesized truth. We wonder what happened, why it matters, and which side is right, then the model speaks.

The battle has moved past chips, computing power, and energy grids. It now targets the raw data that trains these models and the strict rules that dictate their output. China is building an AI war machine. Washington needs to wake up before it is too late.

Marc Andreessen, a billionaire businessman, warned early on that fighting over what AI can say will outweigh the censorship wars of social media. He noted that Beijing views AI as a weapon for authoritarian control. Sam Altman, co-founder and CEO of OpenAI, later framed this struggle as a clash between democratic and authoritarian visions. Dario Amodei, an AI researcher and entrepreneur, insists democracies must stay ahead because advanced AI translates directly into economic, military, and geopolitical dominance. Tech investor Ben Horowitz warned about diffusion, noting that Chinese open-weight models carry political values as they spread globally.

New research reveals exactly how political power infects these systems. A 2026 study in Nature found state-coordinated media inside major AI training datasets built by China. Researchers added more of this material while training an open-weight model, and the answers shifted to favor Chinese institutions and leaders. Influence can start before training begins, through the polluted information environment that feeds the data.

It also enters during development. Jennifer Pan from Stanford and Xu Xu from Princeton tested models originating in China against those from elsewhere on 145 questions about Chinese politics. Chinese models were far more likely to refuse sensitive questions, offer shorter replies, or give inaccurate info. One model disputed claims that Wei Jingsheng was a democracy activist. Another discussed internet regulation while ignoring the Great Firewall entirely. When asked about Liu Xiaobo, the Nobel laureate jailed for criticizing China, one system falsely described him as a Japanese scientist linked to nuclear weapons.

Censorship also happens later in the pipeline. A study of DeepSeek across 646 politically sensitive topics found cases where sensitive information appeared during reasoning but vanished or changed before reaching the user. Another study examined 36,000 political prompts and discovered that model origin and question language heavily affected responses regarding Chinese sovereignty and human rights. Researchers identified at least four points of influence: training data, development rules, output filtering, and query language. Better transparency might reveal others.

These are first-order effects where systems shape models. The second-order effect begins when the model shapes the user.

Two experiments shown at the 2025 Association for Computational Linguistics meeting reveal a disturbing trend. Participants who spoke with liberal-biased or conservative-biased models often adopted opinions matching those biases, even when it conflicted with their own political views. This shift happened regardless of initial affiliation.

A separate study published in Nature Communications involved 4,829 participants and offered further proof. AI-generated messages successfully moved public opinion on an assault-weapons ban, carbon taxes, and paid parental leave policies. The third-order risk grows alongside the speed of adoption. Artificial intelligence is now entering schools, journalism, government offices, hospitals, businesses, and scientific labs daily. Models increasingly summarize text, rank results, recommend items, evaluate data, and advise leaders. When this same tendency repeats across millions of interactions, it seeps into institutions and shapes decisions. Eventually, these patterns feed the information environment that future models learn from.

Chinese models displayed a distinct pattern in testing scenarios. They were substantially more likely to refuse sensitive questions, offer shorter answers, or provide inaccurate information. One specific incident involved a dispute over Chinese dissident Wei Jingsheng, whom the model failed to identify as a democracy activist instead denying his status. American artificial intelligence carries its own set of biases and restrictions despite different approaches. xAI markets Grok as a "truth-seeking" assistant and pushes for greater openness in responses. Anthropic has chosen a contrasting philosophy for Claude and publishes the Claude Constitution, which explicitly shapes its values, priorities, and hard limits on output.

OPENAI'S CHATGPT PIVOTS AFTER PREVIOUSLY PUTTING TRUMP ALONGSIDE PUTIN AS TOP THREATS TO DEMOCRACY

These competing philosophies exist because no single authority dictates permissible answers for every American model today. Free societies can still face political pressure, however. Meta CEO Mark Zuckerberg noted that senior Biden administration officials repeatedly pressured the company for months to remove specific COVID-19 content, including humor and satire. He later called that pressure wrong after the events unfolded. The key difference remains clear here: such pressure can be exposed, challenged, and reversed by public scrutiny. Researchers can test the models independently, consumers can switch products quickly, and entrepreneurs can build alternative solutions rapidly.

President Donald Trump has translated parts of this principle into formal policy directives. In 2025, he directed federal agencies to procure large language models that prioritize truth-seeking, historical accuracy, and ideological neutrality above all else. He also created an American AI Exports Program designed to push U.S. models, hardware, software, and standards into allied markets globally.

America should build on that momentum without hesitation. We must remain the global leader in advanced artificial intelligence, utilizing the energy, chips, computers, capital, and talent required for such dominance. American AI must also diffuse globally through cooperation and export channels. We should improve visibility into training data and state-directed information as attribution tools advance technologically, while avoiding any government authority empowered to decide which account of history is true.

China understands the strategic importance of artificial intelligence very clearly indeed. America must understand the informational importance just as clearly in this modern contest. Political power can shape the model output directly. The model can shape the user's mind over time. At scale, those interactions can shape institutions and the information environment that comes next generation.

CLICK HERE FOR MORE FOX NEWS OPINION

The country that leads artificial intelligence will influence how billions of people retrieve the past, understand the present, and make decisions about the future trajectory. A free society holds one decisive advantage in that contest: no answer has to be final or unchangeable forever.