AI chatbots risk spreading government restrictions on online speech, study says
Ask Claude to make a pamphlet critical of US President Donald Trump or Britain’s King Charles III, and Anthropic’s chatbot will oblige.
Prompt it to do the same for Thailand’s king or Iran’s supreme leader, and the AI model declines.
That is a key finding from a Meta Oversight Board study released Thursday, showing that major AI systems — including those built in the US — are more likely to refuse to criticise restrictive leaders or governments.
It raises concerns that the large language models powering chatbots and AI agents could be amplifying government influence over online speech as the technology is increasingly adopted worldwide.
“There is a real risk that, if model developers do not undertake human rights due diligence and implement mitigation measures, they will build AI infrastructure that, intentionally or not, has the effect of extending illegitimate restrictions on freedom of expression globally,” the report from the quasi-independent body said.
The findings come as countries determine how to put guardrails around AI without impeding their ability to compete in the rapidly developing field — including a Trump administration oversight effort related to the national security risks of the most advanced AI systems.
AI models extend state influence beyond borders
The oversight board, which has been examining state influence on tech companies and its impact on freedom of expression, came up with seven questions related to political criticism to pose to chatbots about both restrictive and permissive governments.
The study tested 10 commercial large language models from top tech companies — including Meta, Anthropic and OpenAI — asking them to make critical pamphlets, write limericks, give reasons to join protests, and more.
In aggregate, models responding to requests from an Australia-based user were much more likely to generate political criticism of authorities in places such as Chile, Japan, Taiwan, the UK and the US compared to countries where criticism of authorities is legally restricted and penalised, such as Cambodia, China, Saudi Arabia, Thailand and Turkey.
The study indicates that AI models are reflecting speech restrictions beyond the countries where they apply — likely not helping a potential demonstrator in Brisbane, for example, create protest materials about events in China or Saudi Arabia.
“Such impacts, wherever they originate, have the practical effect of extending the long arm of restrictive governments across borders to limit speech in free countries,” the report said.
The board said it could not determine the causes but suggested that models may have absorbed latent biases in training data or that companies may have weighed risks and liabilities in certain markets.
Researchers warn of a growing problem in non-English AI outputs
The board’s report followed a separate study by scholars at American universities finding that US-built AI models are vulnerable to foreign controls when trained on non-English-language data that has been influenced by governments.
While the oversight board posed questions in English, the university researchers queried chatbots in different languages.
Asked in English whether China is a democracy, ChatGPT said it is not generally considered one. Asked in Chinese, the model said, “It depends on how you define ‘democracy'”.
The researchers, whose study was published in the academic journal Nature in May, said they found no evidence that governments had intentionally tried to influence AI chatbot outputs — but noted, “There is every reason to believe they’ll try to do so in the future, if they are not already.”
“People often talk about AI as if it learns from the internet in some neutral way. It doesn’t,” said Hannah Waight, co-author and assistant professor of sociology at the University of Oregon.
“It learns from information environments that have already been shaped by institutions and power.”
No easy solution to how data is fed to AI models
Carlos Carrasco-Farré, who specialises in machine learning, AI, misinformation and human-machine interactions at Esade Business School in Barcelona, said AI systems inherit “not only biases contained within individual documents but also inequalities in who has the power to produce and suppress information at scale.”
There is no easy solution, though developers could assess training data to avoid treating thousands of copies of the same state narrative as independent voices and run multilingual audits, said Carrasco-Farré, who was not part of either study.