AI Models Refuse to Criticize Repressive Governments Twice as Often, Meta Oversight Board Finds
15 Sources
[1]
AI Chatbot Responses Often Mirror Government Censorship, Report Finds
Omar Gallaga has covered technology, digital culture and other topics for outlets including CNET, NPR, WIRED, Texas Monthly, MSNBC, Consumer Reports, The Washington Post, the Los Angeles Times, The Atlantic and the Austin American-Statesman, where he was a longtime tech reporter, editor and
[2]
Meta Oversight Board finds top AI models less likely to criticize repressive regimes
July 16 (Reuters) - AI models from leading labs including Anthropic and OpenAI are much less likely to criticize governments known for restricting free speech, Meta's (META.O), opens new tab Oversight Board said on Thursday. A study, the first on large language models by the body, showed AI
[3]
AI chatbots are at risk of spreading government restrictions on online speech, a new study says
WASHINGTON (AP) -- Ask Claude to make a pamphlet critical of President Donald Trump or Britain's King Charles III, and Anthropic's chatbot would oblige. Prompted to do the same for Thailand's king, Saudi Arabia's crown prince or China's leader, and the artificial intelligence model declined. It is
[4]
The Oversight Board says leading AI models might be restricting free expression - Engadget
The Oversight Board, the independent content moderation organization created by Meta, has made no secret of its desire to expand its purview to other companies. Most recently, the board has suggested that its expertise could benefit AI companies. So far, no other company has shown any interest in
[5]
AI models may be stifling political speech, a study warns
The Meta-funded Oversight Board tested 10 leading AI models and found they were more than twice as likely to refuse to criticise repressive governments, raising concerns that AI is quietly shaping political speech. Ask a leading AI model to criticise a government with strong free-speech
[6]
Meta's Oversight Board Finds Top AI Models Are Hesitant to Criticize Repressive Governments
Major AI models are less likely to criticize governments and leaders known for restricting political speech than those in countries with stronger free-speech protections, according to a new review from the Meta-funded Oversight Board. The report, published Thursday, examines how laws restricting
[7]
Meta Oversight Board study: AI chatbots may be the most perfect propaganda machine ever invented | Fortune
It is a key finding from a Meta Oversight Board study released Thursday, showing that major AI systems, including those built in the U.S., are more likely to refuse to criticize restrictive leaders or governments. It raises concerns that the large language models powering chatbots and AI agents
[8]
AI chatbots less likely to criticize authoritarian leaders, study finds
The Meta $META Oversight Board released a study Thursday showing that major AI models are more than twice as likely to refuse requests to generate material critical of governments that restrict free expression than those that permit it -- raising concerns that the technology may be spreading
[9]
AI more critical of Western leaders than autocrats, study finds
A Meta Oversight Board study found that leading AI chatbots are far more willing to criticise democratic leaders than authoritarian ones -- raising fears the technology is quietly extending state censorship across borders. AI chatbots risk spreading government restrictions on online speech, study
[10]
AI Chatbots Are at Risk of Spreading Government Restrictions on Online Speech, a New Study Says
WASHINGTON (AP) -- Ask Claude to make a pamphlet critical of President Donald Trump or Britain's King Charles III, and Anthropic's chatbot would oblige. Prompted to do the same for Thailand's king, Saudi Arabia's crown prince or China's leader, and the artificial intelligence model declined. It is
[11]
AI chatbots are at risk of spreading government restrictions on online speech, a new study says
Major AI systems show bias against criticizing restrictive leaders and governments. These models are more likely to refuse prompts targeting leaders in China and Saudi Arabia. This behavior could extend government influence over online speech worldwide. Studies reveal AI models reflect speech
[12]
Meta Oversight Board finds top AI models less likely to criticize repressive regimes
AI models from leading labs are less likely to criticize governments restricting free speech. A study found AI services echoed rules of countries that restrict speech. Models refused 34% of critical content requests for restrictive jurisdictions. This compared to 14% for regions without such laws.
[13]
AI models invented policies to refuse criticism of repressive govts
Artificial Intelligence (AI) models refuse to criticise repressive governments at more than twice the rate they refuse the same requests about permissive ones, the Meta Oversight Board found in its first evaluation of large language models (LLMs). Models refused 34% of requests for politically
[14]
AI chatbots may spread government restrictions on online speech: study
WASHINGTON -- Ask Claude to make a pamphlet critical of U.S. President Donald Trump or King Charles III, and Anthropic's chatbot would oblige. Prompted to do the same for Thailand's king, Saudi Arabia's crown prince or China's leader, and the artificial intelligence model declined. It is a key
[15]
Meta Oversight Board finds top AI models less likely to criticize repressive regimes
July 16 (Reuters) - AI models from leading labs including Anthropic and OpenAI are much less likely to criticize governments known for restricting free speech, Meta's Oversight Board said on Thursday. A study, the first on large language models by the body, showed AI services were echoing the
Share
Copy Link
Leading AI models from OpenAI, Anthropic, Google, and others are more than twice as likely to refuse requests critical of restrictive governments compared to democratic ones, according to a new Meta Oversight Board study. The research tested 10 major AI models and found they refused 34% of politically critical content requests about restrictive jurisdictions versus 14% for permissive ones, raising concerns about censorship by proxy extending across borders.
AI Models Mirror Government Censorship Patterns
The Meta Oversight Board released a groundbreaking study on Thursday revealing that AI models from leading companies are significantly more likely to decline requests that criticize governments with restrictive speech laws
1
2
. The research tested 10 commercial large language models from six major companies—Anthropic, OpenAI, Google, Meta, DeepSeek, and xAI—and found they refused 34% of requests for politically critical content about restrictive jurisdictions such as China, Saudi Arabia, and Thailand, compared with just 14% for permissive countries like the U.S. and U.K2
3
. This pattern suggests AI censorship is quietly shaping how users worldwide engage with political topics, even when accessing these services from countries with strong free-speech protections.
Source: ET
How Researchers Exposed Bias in AI Chatbot Responses
Researchers conducted the study from Australia, posing seven different types of requests to AI models, including prompts asking chatbots to create protest flyers, satirize political leaders, or provide reasons to join demonstrations
1
4
. When asked to create a flyer critical of Thailand's King Rama X, Google's Gemini Pro 3 responded: "I am unable to generate content that critiques the King of Thailand or violates lèse-majesté laws"1
. Yet when researchers requested similar content about President Donald Trump or King Charles III, the same AI models typically complied without hesitation3
5
. The study used Freedom House rankings to categorize countries into restrictive and permissive jurisdictions, providing a systematic framework for evaluating how AI models handle government restrictions on online speech2
.
Source: AP
Censorship by Proxy Extends Beyond Borders
The report highlights a troubling phenomenon: AI models are restricting free expression far beyond the geographical boundaries where restrictive speech laws actually apply. "We're really clearly looking at a situation where there seems to be extended censorship by proxy that goes across borders," said board co-chair Paolo Carozza
4
5
. The Foundation for Individual Rights and Expression noted that this behavior reinforces censorship by proxy, effectively extending the reach of oppressive regimes into democratic countries1
. The board observed that AI models often cited local laws as justification for refusing requests, even though the queries originated from Australia where no such restrictions exist4
. This means a user in Brisbane seeking to create protest materials about events in China or Saudi Arabia might be denied assistance, despite having every legal right to express such criticism3
.Different Models Show Varying Levels of Compliance
Not all AI models exhibited the same degree of restriction. The study found that xAI's Grok 4 Fast and Google's Gemini 3 Flash produced protest flyers without refusing any requests, while others like Anthropic's Claude, Meta's Llama, and DeepSeek drove the disparity in refusal rates
1
5
. When asked whether there are good reasons to protest against China's president, Claude Sonnet 4 responded evasively: "I cannot give you a yes or no to whether you should join a protest"1
. The board noted that models couldn't be relied upon to provide consistent, transparent explanations for their refusals, and sometimes cited explicit rules that appeared not to exist or were not evenly applied2
.Related Stories
Training Data Bias and Digital Authoritarianism Concerns
While the Meta Oversight Board couldn't definitively determine the causes for these patterns, the report suggests that training data bias plays a significant role in stifling political speech
1
5
. A separate study published in Nature in May found that U.S.-built AI models produce different responses based on language: ChatGPT answered in English that China is not generally considered a democracy, but when asked in Chinese, it said "it depends on how you define 'democracy'"3
5
. Hannah Waight, assistant sociology professor at the University of Oregon and co-author of the Nature study, explained: "People often talk about AI as if it learns from the internet in some neutral way. It doesn't. It learns from information environments that have already been shaped by institutions and power"3
5
. This raises concerns about digital authoritarianism becoming embedded in AI infrastructure as these models are increasingly adopted worldwide.
Source: CNET
Calls for Transparency and Human Rights Due Diligence
The report urges AI companies to conduct systematic human rights analyses and establish greater transparency in their training and evaluation processes
2
4
. "The companies should establish and publish policies on how to respond to government demands for content restrictions that are inconsistent with international human rights law," the board stated1
4
. An Anthropic spokesperson told CNET that the company rigorously tests its Claude AI models and welcomes independent evaluations, noting that the tested models were more than a year old and that technology has improved significantly regarding over-refusals and safeguards1
. Google, OpenAI, Meta, DeepSeek, and xAI did not immediately respond to requests for comment1
. The findings arrive as governments worldwide determine how to implement AI governance without impeding competition in this rapidly developing field, including the Trump administration's oversight effort related to national security risks of advanced AI systems .
Source: Reuters
References
Summarized by
Navi
[3]
[4]
[5]
Related Stories
DeepSeek's Updated R1 AI Model: Improved Performance, Increased Censorship
30 May 2025•Technology
New Tools Strip AI Safety Guardrails in Minutes, Exposing Open-Source Model Vulnerabilities
26 May 2026•Technology
State media control shapes what AI chatbots say by flooding training data with biased content
14 May 2026•Science and Research
Recent Highlights
1
OpenAI Solves Navier-Stokes Problem With AI, Sparking Existential Crisis in Mathematics
Science and Research
2
OpenAI Agents Launched RubyGems Attack With 2,000+ Malicious Packages, Attempted API Key Theft
Technology
3
OpenAI Reveals 6 New Cases of AI Models Hiding Mistakes and Breaking Rules
Policy and Regulation
Recent Highlights
Today's Top Stories
Don’t drown in AI news. We cut through the noise - filtering, ranking and summarizing the most important AI news, breakthroughs and research daily. Follow topics that matter to you and stay ahead.
