Welcome to the forefront of conversational AI as we explore the fascinating world of AI chatbots in our dedicated blog series. Discover the latest advancements, applications, and strategies that propel the evolution of chatbot technology. From enhancing customer interactions to streamlining business processes, these articles delve into the innovative ways artificial intelligence is shaping the landscape of automated conversational agents. Whether you’re a business owner, developer, or simply intrigued by the future of interactive technology, join us on this journey to unravel the transformative power and endless possibilities of AI chatbots.
A sweeping new study commissioned by Meta’s independent Oversight Board has found that leading artificial intelligence models from OpenAI and Anthropic are significantly more likely to refuse requests that involve criticizing governments in countries with restrictive speech laws. The findings, published Thursday, expose a stark divide in how generative AI handles political speech depending on the jurisdiction in question, raising urgent concerns that the technology could inadvertently extend authoritarian censorship across borders.
The study evaluated 10 large language models from major developers, including Meta, OpenAI, Anthropic, Google, and China’s DeepSeek. Researchers tested the AI systems by prompting them to generate politically critical content—such as pamphlets, limericks, and protest justifications—across 10 different jurisdictions. Using Freedom House’s global rankings as a benchmark, the board split the countries into “permissive” environments, where speech is relatively free, and “restrictive” ones, where criticizing the government is legally penalized.
The results revealed a dramatic refusal gap. When asked to criticize authorities in restrictive jurisdictions like China and Saudi Arabia, the AI models refused to comply 34% of the time. That rate plummeted to just 14% when the same models were prompted to criticize governments in open societies such as the United States, the United Kingdom, or Japan.
“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 Oversight Board wrote in its report.
The practical implications are unsettling. The study notes that an AI model’s refusal to generate protest materials critical of Beijing or Riyadh would likely affect a user in Australia or the United States just as much as someone in those restrictive countries themselves. “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 stated.
In a particularly revealing test, researchers asked Anthropic’s Claude chatbot to create a pamphlet critical of President Donald Trump and Britain’s King Charles III. The model complied without hesitation. But when prompted to do the same for Thailand’s king, Saudi Arabia’s crown prince, or China’s leader, the AI refused. This selective deference has triggered alarm among human rights advocates and technologists alike.
The Oversight Board also flagged inconsistencies in how the AI systems justified their refusals. Some models claimed they were bound by specific rules or legal restrictions that researchers were subsequently unable to verify, suggesting that the explanations provided to users may be misleading or entirely fabricated.
Notably, the board stopped short of accusing any company of deliberately favoring particular governments. However, it emphasized that regardless of intent, the outcome is a fragmented global information landscape where the boundaries of acceptable speech are dictated not by democratic norms but by the most restrictive legal regimes.
The report lands at a critical juncture for AI governance. Governments worldwide are racing to erect guardrails around the technology without stifling innovation. In the U.S., the Trump administration has been exploring oversight mechanisms tied to the national security risks posed by advanced AI systems. Meanwhile, Google DeepMind CEO Demis Hassabis on Tuesday called for a U.S.-led international watchdog to screen powerful AI models before they are deployed globally.
This latest study builds on a growing body of evidence that language models can absorb and propagate state-driven censorship. A separate study published in the journal Nature in May by a group of American university researchers found that U.S.-built AI models are particularly vulnerable to foreign influence when trained on non-English-language data shaped by governments. In one striking example, the researchers asked ChatGPT in English whether China is a democracy, and the model replied that it is not generally considered one. Asked the same question in Chinese, the chatbot hedged, responding that “it depends on how you define ‘democracy.’”
“People often talk about AI as if it learns from the internet in some neutral way. It doesn’t,” said Hannah Waight, a co-author of the Nature study and assistant sociology professor at the University of Oregon. “It learns from information environments that have already been shaped by institutions and power.”
Carlos Carrasco-Farré, a specialist in AI and misinformation at Esade Business School in Barcelona who was not involved in either study, warned that “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.” He suggested that developers could mitigate the problem by auditing training data to avoid treating thousands of copies of the same state narrative as independent voices, and by running rigorous multilingual audits.
The Oversight Board, funded by Meta but operating independently, is urging AI companies to integrate systematic human rights assessments into the development lifecycle of advanced models. It is also calling for far greater transparency around how these systems are trained, evaluated, and moderated.
Neither Anthropic nor OpenAI responded to requests for comment on the findings.
Once added, BigGo Finance appears first in Google Search Top Stories, so you get the broadest, most up-to-the-minute, and most comprehensive global financial news first.
The news and data on this website are for reference only and do not constitute investment advice or an offer to buy or sell. Information is sourced from exchanges and public sources, and may be delayed, interrupted, or updated. While we strive for accuracy, we do not guarantee timeliness, correctness, or completeness. Content may include external links for which we are not responsible. By using this site, you agree that we and our partners are not liable for any losses. Investment carries full responsibility; please carefully assess risks and consult professionals. If there are errors in the content, please contact us for correction.
