Polling Shows Europeans Want Legal Requirements for AI Development in the AI Act

European polling by YouGov has revealed decisive support amongst the most populous EU member states for the EU AI Act to regulate foundation models and their creators. These findings arrive at a time when there are increasingly aggressive efforts by France, Germany and Big Tech to lobby for foundation models to be exempted from the Act.The poll, run in Germany, France, Italy & Spain, found decisive majorities in favour of the EU AI Act regulating AI model creators. In all four nations:Over 80% of respondents believe that the EU AI Act should hold AI model creators either wholly or partially responsible for harms resulting from their modelsOver two thirds of respondents agreed that the EU AI Act should require the creators of foundation models to prove that their models are safe

In none of the polled countries did more than 6% of respondents claim that only the distributors of AI models should be held legally responsible for harms resulting from the models. This is a distinct contrast to attempts by French politicians and AI companies to place the Act’s regulations solely upon downstream AI products, rather than on the developers of foundation AI models. There is robust transnational support for the EU AI Act to place the burden of proof upon foundation model creators to prove that their models are safe, with large majorities (including 78% of Spaniards) believing that the EU should require foundation models creators tor prove that their models are safe.There is also broad support for government regulation of powerful AI. In all four countries:More than two thirds of respondents supported governments having the power to disable powerful AI systems if they become dangerous More than 63% of respondents agreed that governments should require AI companies to register training runs for powerful modelsOver 60% claim they do not trust technology CEOs to act in the public’s interest with regards to AI

The strong support for requiring technology companies to register large training runs with governments shows how public opinion diverges from tech lobbyists on how to regulate AI. If, as some French politicians and Big Tech companies insist, only downstream AI applications can be regulated, the developers of foundation models would be free to train their systems without limit or oversight. Yet support for government oversight of large training runs ranges from 63% in France to 68% in Germany.Trust in tech CEOs to act in the public’s interest with regards to AI is consistently low across all four countries. In the country with the highest level of faith in tech CEOs, Italy, only 28% claim they trust technology companies.  France is the nation most distrustful of Big Tech, with only 19% saying that they trust tech CEOs and 70% saying they distrust. This is remarkable, given that the American-funded AI company Mistral is at the forefront of France’s campaign to exempt foundation models from the EU AI Act. 

Respondents from all nations showed strong support for governments having the power to disable powerful AI systems if they prove to be dangerous. Across all polled electorates, over two thirds supported governments having such power - including 76% of Spanish respondents.These results show that European public opinion is strongly on the side of regulating the creation and development of powerful AI systems, including foundation models. This is in direct contrast to campaigns from Big Tech to regulate ‘the usage of the technology, not the technology itself’.

See UK Poll results here.

All figures, unless otherwise stated, are from YouGov Plc.  Total sample size was 1,095 adults from France, 1136 from Germany, 1110 from Spain, 1052 from Italy. Fieldwork was undertaken between 21st - 28th November 2023.  The survey was carried out online. Each survey sample has been weighted and is representative of all adults in its respective country (aged 18+).