The article discusses how the United States and China are competing to shape the development of artificial intelligence (AI), which is becoming a primary source for information about history, politics, science, and culture. It highlights concerns raised by figures like Marc Andreessen, Sam Altman, Dario Amodei, and Ben Horowitz that this contest could determine whether AI promotes democratic or authoritarian values globally. The piece cites research indicating that Chinese state-coordinated media can influence major AI training datasets, leading to biased outcomes in responses related to political topics.
Written by the local model on 2026-09-14,
using this article's own text rather than the other coverage of the
same event (that is the story summary below).
As artificial intelligence (AI) becomes how we learn and understand the past, America and China are competing over systems that will shape what billions believe.
asserted
billions → become → what
A library gave you 10 books.
asserted
library → give → books
Google gave you ten links.
asserted
Google → give → links
Artificial intelligence (AI) increasingly gives you one answer.
asserted
intelligence → give → answer
Generative AI is becoming the layer between us and the record itself.
asserted
AI → become → us
It reads and synthesizes information before we see it.
asserted
we → read → it
We ask what happened, what it means and whom to believe, and the model answers.
asserted
model → ask → whom
As billions of people use AI to encounter history, politics, science and culture, the contest between the United States and China extends beyond chips, compute and energy to the information from which AI learns and the rules that shape what it tells us.
asserted
it → use → us
Billionaire businessman Marc Andreessen argued early that the fight over what AI is allowed to say could become more consequential than the censorship battles of the social-media era and warned that China sees AI as a tool of authoritarian control.
uncertain
China → argue → control
OpenAI co-founder and CEO Sam Altman later described the contest as one between democratic and authoritarian visions of AI.
asserted
Altman → describe → AI
AI researcher and entrepreneur Dario Amodei has argued that democracies must maintain leadership because advanced AI will translate into economic, military and geopolitical power.
asserted
AI → argue → power
Tech entrepreneur Ben Horowitz has focused on diffusion, warning that Chinese open-weight models can carry political assumptions and values as they spread around the world.
asserted
they → focus → world
The research now shows several ways political power can enter an AI system.
asserted
power → show → system
A 2026 Nature study found Chinese state-coordinated media inside major AI training datasets.
asserted
study → find → datasets
Researchers added more of that material while training an open-weight model, and its answers became more favorable toward Chinese political institutions and leaders.
asserted
answers → add → institutions
Influence can begin before training, through the information environment that supplies the data.
asserted
that → begin → data
It can also enter during development.
asserted
It → enter → development
Jennifer Pan of Stanford and Xu Xu of Princeton tested China-originating and non-China-originating models on 145 questions about Chinese politics.
asserted
Pan → test → politics
Chinese models were substantially more likely to refuse sensitive questions, give shorter answers or provide inaccurate information.
asserted
models → refuse → information
One disputed that Chinese dissident Wei Jingsheng was a democracy activist.
asserted
Jingsheng → dispute → ?
Another discussed internet regulation while omitting China’s Great Firewall.
asserted
Another → discuss → Firewall
Asked about Nobel laureate Liu Xiaobo, who was jailed for criticizing China, one falsely described him as a Japanese scientist associated with nuclear weapons.
asserted
one → ask → weapons
Other research finds censorship later in the pipeline.
asserted
research → find → pipeline
A study of DeepSeek across 646 politically sensitive topics found cases in which sensitive information appeared during the model’s reasoning and was then omitted or reformulated before the answer reached the user.
asserted
answer → find → user
A separate study of 36,000 political prompts found that model origin and the language of the question affected political responses, particularly on Chinese sovereignty and human rights.
asserted
origin → find → sovereignty
The research identifies at least four points of influence: training data, development rules, output filtering and the language of the query.
asserted
research → identify → query
Better transparency may reveal others.
uncertain
transparency → reveal → others
These are first-order effects: political systems can shape the model.
asserted
systems → shape → model
The second-order effect begins when the model shapes the user.
asserted
model → begin → user
In two experiments presented at the 2025 Association for Computational Linguistics annual meeting, people who interacted with liberal-biased or conservative-biased models were more likely to adopt opinions and decisions aligned with the model’s bias, even when it ran against their own political affiliation.
asserted
it → present → affiliation
A Nature Communications study involving 4,829 participants also found that AI-generated political messages shifted attitudes on an assault-weapons ban, carbon taxes and paid parental leave.
asserted
messages → involve → ban
The third-order risk comes with scale.
asserted
risk → come → scale
AI is moving into schools, journalism, government, medicine, business and scientific research.
asserted
AI → move → schools
Models increasingly summarize, rank, recommend, evaluate and advise.
asserted
Models → summarize → ?
The same tendency repeated across millions of interactions can work its way into institutions and decisions, and eventually into the information environment future models learn from.
asserted
models → repeat → environment
Chinese models were substantially more likely to refuse sensitive questions, give shorter answers or provide inaccurate information.
asserted
models → refuse → information
One disputed that Chinese dissident Wei Jingsheng was a democracy activist.
asserted
Jingsheng → dispute → ?
American AI has biases and restrictions too.
asserted
AI → have → biases
xAI calls Grok a "truth-seeking" assistant and has pushed toward greater openness.
asserted
xAI → call → openness
Anthropic has chosen a different philosophy for Claude and publishes Claude’s Constitution, which explicitly shapes its values, priorities and hard limits.
asserted
which → choose → values
…and 19 more, not listed.
