The global race to develop artificial intelligence is increasingly becoming a contest over more than computing power and technology. It is also raising questions about how AI systems shape what billions of people believe.
Generative AI is becoming an intermediary between users and the information they seek, synthesising material before presenting an answer.
Researchers cited in a new Fox News opinion article argue that political influence can enter AI systems through training data, development rules, output filters and even the language used in a question.
A 2026 Nature study found Chinese state-coordinated media in major AI training datasets. Researchers who added more of that material when training an open-weight model found that its responses became more favourable towards Chinese political institutions and leaders.
Other research has found differences between Chinese and non-Chinese AI models when answering politically sensitive questions. A study by Jennifer Pan of Stanford University and Xu Xu of Princeton University tested models on 145 questions about Chinese politics and found that Chinese-originating models were more likely to refuse sensitive questions, provide shorter answers or give inaccurate information.
Research has also examined what happens after an AI system produces an answer. Two experiments presented at the 2025 Association for Computational Linguistics annual meeting found that people interacting with politically liberal- or conservative-biased models were more likely to adopt views and decisions consistent with the model's bias, including when those views conflicted with their own political affiliation.
A separate Nature Communications study involving 4,829 participants found that AI-generated political messages could shift attitudes on issues including an assault-weapons ban, carbon taxes and paid parental leave.
The potential impact extends beyond individual users as AI becomes more widely used in schools, journalism, government, medicine, business and scientific research. Models increasingly summarise, rank, recommend and advise, potentially allowing patterns in individual interactions to influence institutions and the wider information environment.
The article argues that American AI systems also have biases and restrictions, while companies such as xAI and Anthropic have adopted different approaches to defining the values and limits of their models.
It says the US should maintain leadership in advanced AI while improving transparency around training data and state-directed information. At the same time, it argues that governments should not be given authority to determine which version of history is true.
The central concern, according to the article, is that political power can influence AI systems, AI systems can influence users, and repeated interactions at scale can ultimately influence institutions and the information environment on which future AI systems are trained.
As AI becomes a primary way for people to retrieve information about the past, understand current events and make decisions about the future, the contest over the technology is increasingly also a contest over the information it delivers.
By Aghakazim Guliyev
