AI Chatbots Risk Reinforcing Censorship: Study

The ChatGPT application icon is displayed on the screen of a smartphone in Chicago on Aug. 4, 2025.

A new study by the Meta Oversight Board has raised concerns that leading artificial intelligence (AI) chatbots may be reinforcing government restrictions on free expression by refusing to generate criticism of political leaders in countries with strict speech controls.

The report found that Anthropic's Claude readily generated content critical of US President Donald Trump and Britain's King Charles III, but declined similar requests involving China's president, Thailand's king and Saudi Arabia's crown prince.

The board warned that AI developers could unintentionally extend illegitimate restrictions on free speech worldwide unless they conduct comprehensive human rights assessments when designing and training their systems.

The study evaluated 10 major AI models, including those developed by Meta, OpenAI and Anthropic, using prompts related to political criticism.

Researchers found that the chatbots were generally more willing to criticise leaders in countries with stronger protections for free expression than authorities in countries where dissent is more tightly restricted.

The findings follow a separate study published in the journal Nature, which reported that AI models trained on non-English data may reflect government-influenced narratives.

That research found that ChatGPT produced different responses about China's democracy depending on whether users asked the question in English or Chinese.

Experts said AI systems inevitably inherit biases from the data used to train them and urged developers to strengthen multilingual testing, conduct regular bias audits and reduce the influence of repeated state-backed narratives in training datasets.