Anthropic applies EU AI-generated content marks worldwide
About 1 in 5 U.S. voters have asked AI chatbots about elections and politics, according to a recent survey cited by Jordi Calvet-Bademunt. Calvet-Bademunt, a Vanderbilt University research fellow and visiting scholar of political science who studies free speech policy, used the survey finding to examine how government rules affect what chatbot users hear about political issues.
In two reports published in 2025 and 2026, Calvet-Bademunt and his colleagues worked with global legal experts to examine AI regulation in 12 countries and regions. Their assessment found that governments often used national security or anti-state propaganda laws rather than new AI-specific legislation to control what models could say. Calvet-Bademunt also said many technology companies apply the same policies worldwide, so answers received by American voters may be shaped by rules at home and abroad.
He wrote that AI can cause real harm, including launching cyberattacks and generating sexualized images of children, and said governments should address those dangers. His assessment was that some governments use those legitimate safety concerns as a pretext to limit political discussion, especially chatbot responses about people in power.
Calvet-Bademunt cited a 2024 case in which an Indian journalist asked Google’s Gemini whether Narendra Modi’s policies were fascist; Gemini responded that some experts said yes. The Indian government then required official permission before AI models described as “unreliable” could be released, but authorities later dropped the requirement. Calvet-Bademunt said terms such as “unreliable” and “under-testing” can be used for political purposes.
Calvet-Bademunt also wrote that China requires AI models to uphold “socialist core values.” In Turkey, a court blocked posts by Grok, the chatbot built into the social platform X, that allegedly insulted President Recep Tayyip Erdoğan, founding father Mustafa Kemal Atatürk and religious values. Turkish law scholar Yaman Akdeniz identified the Turkish rulings among the earliest documented cases of state censorship of chatbots.
He also cited Vietnam, where a court sentenced two activists living in Germany to 17 years in prison each in December 2025 for spreading anti-state material. Prosecutors accused the activists of using AI to fake voices and images of Vietnamese leaders. Calvet-Bademunt said the case illustrates how AI-related prosecutions can target political criticism.
Some of the government measures Calvet-Bademunt described apply only to people residing in a particular country, while others cross borders. The European Union’s AI Act, adopted in 2024, requires any company offering AI tools in Europe, wherever it is based, to mark AI-generated content. Anthropic applies those marks globally because it has no reliable way to apply them by region. Calvet-Bademunt described that reach as the Brussels Effect, in which European rules end up applying elsewhere.
Calvet-Bademunt said European rules can also affect what chatbots say about politics. In August 2026, the European Union placed ChatGPT under the strictest tier of its Digital Services Act, described as Europe’s online safety rulebook. OpenAI, which created ChatGPT, must assess and reduce the service’s “systemic risks,” including election risks such as disinformation.
The author reported that the Dutch Data Protection Authority tested four chatbots before the 2025 national election and found that the political middle ground “virtually disappears” in their voting advice. The authority urged providers to “prevent their systems from being used for voting advice.”
Calvet-Bademunt said European rules can also be used to pressure platforms. In 2024, Thierry Breton, then the European Union’s top digital rules enforcer, reminded Elon Musk of his legal obligations before Musk’s live interview on X with then-presidential candidate Donald Trump. Breton warned against the spread of “harmful content.”
Calvet-Bademunt also examined U.S. rules. A 2025 executive order by President Donald Trump titled “Preventing Woke AI in the Federal Government” requires AI models purchased by the federal government to be “truth-seeking,” ideologically neutral and not favor ideas such as “diversity, equity and inclusion.”
Calvet-Bademunt said “truth-seeking” can carry different meanings. As an example, he cited Trump’s recent ban of three news outlets from the White House for reporting he characterized as “FICTION and LIES,” along with the president’s dismissal of media polls as “fake.” More recently, the Federal Trade Commission suggested that AI companies steering answers toward undisclosed ideological or biased objectives might break consumer protection law.
A 2025 Pew Research Center survey found that 8 in 10 Americans believe Republican and Democratic voters cannot agree on basic facts. Calvet-Bademunt used the result to argue that government mandates to seek truth are often driven by ideological belief systems, particularly on political issues.
He also noted that dozens of U.S. states have passed laws restricting AI-generated political deepfakes. California and Hawaii banned sharing deceptive AI content about candidates in the months around an election, but U.S. courts blocked both laws, citing the First Amendment.
For voters, Calvet-Bademunt advised treating chatbot responses as a starting point. He said answers may be shaped by governments at home and abroad and should not replace critical thinking at the ballot box.