Try the Grand Continent. The ideas and analyses you won’t find anywhere else. But you’ll see them everywhere soon. In your inbox starting at 8 euros per month
Artificial intelligence is at a tipping point. The technology is advancing at an ever-faster pace, the valuation of sector companies is reaching record levels, and it now accounts for the bulk of the United States’ GDP growth. At the same time, public opinion is becoming more resistant, and policymakers, who were traditionally more favorable, are turning. AI populism is beginning to take shape clearly. Concern is rising about the upheavals AI is bringing.
Perceptions, however, are far from settled. Users value the tool personally and are adopting it en masse. Politicians and experts who call for more regulation do not agree on the type of oversight to implement. Thus, the politicization of the topic is advancing, but the fault lines remain fluid, which means that the loss of trust is by no means irreversible.
What are the terms of this debate today? Is there a path to reverse the negative narrative around AI?
1 — Widespread Adoption, but Negative Perception
The adoption of AI is very high, both personally and professionally. Seventy-one percent of the public report having used at least one AI tool in the last six months. Among these users, 92% use it as an alternative to search engines to inform themselves and conduct research. 61% report using it in a professional context and 48%, including two-thirds of young adults, use it as emotional and relational support.
However, when asked about their opinion, positive or negative, of several commonly used technologies, AI ranks last, with 18% positive opinions, compared with 78% for search engines, and 74% for smartphones. Even social networks fare better, with 35% positive opinions.
Politically, the topic is beginning to become divisive, especially in the United States, where Republicans hold a more favorable view of AI (net sentiment +44%) than Democrats (+26%). But the most determining explanatory factor is not so much ideological orientation as voter’s propensity to vote: abstainers are markedly less enthusiastic than others.
Across all studied countries, the youngest adults, those with the highest incomes and the most developed digital skills are the most favorable to AI, as are those who already use these tools. AI’s image could thus improve as adoption grows, but could also continue to degrade if deployment negatively affects its strongest supporters, notably the young and the most educated.
2 — Rising Concern Among Policymakers
The experts we interviewed say they are more worried than six months ago, notably policymakers, who express increasing fears about AI risks, particularly the loss of human oversight over systems.
With caveats. The deterioration is clear in the United States, while the experts surveyed in Japan share a more confident view of the technology’s potential and the government’s ability to regulate it pragmatically.
Gaps also widen between political actors and industry players: the latter, while not homogeneous in their positions, focus more on AI-driven productivity gains and the sector’s commercial dynamics.
How to explain AI’s degraded image? Several European experts interviewed consider that the problem lies less in the technology itself than in the reputation of sector companies, their communication strategies, and the profiles of their leaders.
3 — A Critical Look at the US-China Competition
The global AI race does not enjoy broad public support. When asked to choose between AI competitiveness and other priorities such as security, copyright protection, job protection, taxation, or environmental issues, two-thirds of those surveyed do not place competitiveness first.
64% of respondents in the United States believe that their country “should slow down and set the global standard for safe and responsible AI,” while 38% believe it “should lead the world in AI development, even if that means accepting some risks along the way.” In other countries, developing responsible AI is also judged more important (64%) than pursuing competitiveness “above all” (36%).
A majority even thinks that the US-China competition for AI domination will ultimately be harmful to the world, regardless of the winner: 54% in the United States and 52% in other countries. This sentiment is shared by several American experts interviewed who reject the idea of using the rivalry with China as a pretext to slow down AI safety measures.
4 — Support for Sovereignty Policies, While Doubting Their Effectiveness
Public opinion and experts support policies to reduce foreign dependencies in AI. Thus, 76% of those surveyed consider AI sovereignty important, or very important. And 62% of respondents in Europe, Canada, and Japan believe their country should reduce its reliance on China and the United States.
Outside the United States, policymakers and regulators view the government’s decision to restrict the export of AI models as the triggering moment that crystallized a sense of urgency around developing independent AI capabilities.
But backing these sovereignty policies does not mean being convinced of their effectiveness. In Europe, Canada, and Japan, the experts surveyed express little confidence in these countries’ ability to compete with China or the United States in advanced AI models.
In Japan, some officials even question the need to compete at the model level, arguing instead for a sovereignty strategy targeted at certain segments of the AI value chain.
Public opinion shows pragmatism regarding support for local firms. 44% of French and 40% of Japanese people say they are comfortable with foreign technologies, and the majority are also with foreign companies that strengthen their local footprint, notably by paying taxes, creating jobs, and investing locally.
Respondents also support local procurement policies: 60% approve them, even if this leads to higher costs or less capable technology for public services.
Industry experts, however, are more skeptical about local procurement and sovereignty policies in AI. They contend that AI competitiveness and the development of truly globally competitive companies require access to the best technologies, regardless of origin.
5 — Growing Likelihood of AI Regulation, but No Consensus on Content
The question of AI regulation is now on the agenda in each of the seven countries studied. It is, however, more likely in countries that do not yet have it. Yet there is no consensus on the principle of regulation or its modalities. The regulatory fragmentation of AI is likely to persist, or even worsen.
In Europe, some experts, though worried about new safety risks, argue that new regulation is unnecessary: existing regulations are enough, they simply need to be enforced. In countries lacking national AI regulation, such as the United States and the United Kingdom, the interviewed experts now deem the adoption of national legislation and/or international coordination necessary.
Public opinion clearly rejects self-regulation in AI safety: only 7% of respondents express support for industry self-regulation, with no marked preference for a public-control modality over another. The most supported option is to subject new models to government authorization accompanied by independent testing. By contrast, suspending the development of AI models receives less support than expected, with only 13% endorsement.
Regarding the protection of children, a majority of public opinion supports the principle of banning under-16s from using AI-based chatbots and image generators, and rejects the argument that this would hinder their development or impede digital skills acquisition.
Public opinion and experts, however, are divided on the feasibility of such restrictions. A slim majority in the Netherlands believes that a ban on AI use for those under 16 would be “practically impossible to enforce.”
6 — Can We Still Turn the Debate’s Terms Around?
It is not certain that the explanatory work will be carried out successfully. Data centers are a good example. In France, Germany, and Japan, a majority of respondents say they “don’t know what a data center is” (57%, 59%, and 63% respectively). This share is lower in the United States, where the debate over their construction and operation is intense, but still reaches 37%.
Public opinion is pragmatic. When asked about AI’s benefits, everyday-use cases are cited before macroeconomic arguments. Convenience (enabling scheduling, financial management, and quicker access to advice) comes first (29%), ahead of inevitable progress (26%) and universal knowledge (25%). Economic growth lags well behind (12%). There is a hard core of 20% of people hostile to AI and convinced by none of its uses, but 80% of respondents acknowledge the technology’s benefits.
Is there a way to strengthen trust in the sector? The experts we interviewed proposed a wide range of measures the industry could adopt to improve its image. They advocate certification and labeling procedures, transparency mechanisms, and enforceable codes of conduct. Others suggest the industry highlight AI’s benefits in the scientific domain and use cases in sectors such as health. Corporate policies restricting AI use, setting ethical limits, or governing the deployment of models are also welcomed.
It is striking that a common thread runs through all these proposals: the trust deficit in the AI sector. Whatever the case, it will be at the heart of any effort to influence political opinion and actions, whether through regulation or development.