AI Models Have Shown Bias Toward State-Controlled Media
Researchers found that AI models frequently mirror the political narratives of nations with tight media control.
Updated on Sept. 20, 2026 in Artificial Intelligence
Live Poll
Do you trust artificial intelligence models to provide neutral and unbiased political information?
A study from the University of Oregon revealed that AI models exhibit favorable biases toward governments in countries with strong media control when prompted in their native languages. These models appear to absorb political narratives present in their training data.
Why it matters
AI systems learn from the digital information environments they ingest, meaning powerful institutions can inadvertently influence model outputs. This raises concerns about how training data curation impacts the neutrality of AI responses.
Analysis of the Common Crawl-derived dataset identified 3.1 million documents overlapping with Chinese state-controlled media, representing 1.64% of the total dataset. When prompted in Chinese, AI models produced favorable responses regarding political leaders 75.3% of the time.
The players
University of Oregon
A public research university located in Eugene, Oregon, known for its extensive academic research programs across multiple disciplines.
OpenAI
An artificial intelligence research organization that develops large language models utilizing vast datasets like Common Crawl.
The details
Researchers studied 37 countries and found that models frequently favor the government line when users interact in languages prevalent in countries with high media control. While state-controlled media makes up only a small fraction of the training data, its high frequency relative to other local sources significantly tilts model behavior.
Timeline
University of Oregon study findings were published on September 20, 2026.
The Tech Race
This research follows the pattern set by the Common Crawl open-source repository, which acts as the primary data foundation for modern large language models. The findings reveal how the sheer scale of open-web data allows state-sponsored narratives to permeate AI logic, challenging the industry drive toward objective models.
Users interacting with AI in languages other than English may receive information that reflects the dominant political narratives of their region. These findings suggest that responses from current AI models may not always be neutral when navigating topics sensitive to state-controlled institutions.
The takeaway
The findings serve as a reminder that AI models are essentially mirrors of the training data provided to them. Users should remain critical of AI-generated content, especially when navigating topics where media environments are heavily shaped by political entities.
Further reading
For more on the challenges of training data, see our Artificial Intelligence section.
Source note: This article includes information reported by KOIN 6 Portland.
Live Poll
Do you trust artificial intelligence models to provide neutral and unbiased political information?
