POLITICO's poll finds pluralities willing to fall behind on AI for more caution

Support for continued development runs 30% to 40% in each of the six countries, against 45% in Germany and 51% in the United Kingdom for a pause ^[3][2]. Taken as bounds, those ranges put the pause camp ahead by at least 5 points and at most 21 ^[1], and leave between 9 and 25 points of each national sample in neither camp ^[2].

Pluralities in several countries said it is acceptable for their country to fall behind others if that buys more responsible and secure development ^[4]. Whether any of that becomes a rule is a separate question. Trump called the AI warnings a "HOAX" and the prospect of an immediate bipartisan deal fell with it ^[7]. Nvidia chief executive Jensen Huang and former White House adviser David Sacks have both played down the dangers and resisted new regulation ^[8].

Europe already has the rule. The EU adopted the world's first AI regulation in 2024 and hosts almost no providers of frontier models, with France's Mistral the closest contender ^[9]. Most of last year's effort in Brussels went the other way, into preparing billions in compute capacity and lifting low adoption rates ^[12]. Then Ursula von der Leyen used her annual address to say "if the people developing the technology are clear, then we should be too" ^[10], and invited the main AI labs to talk about how the EU could "support" their efforts to pace the frontier ^[11].

"This poll fits the classic techno-panic pattern: few harms today, plenty of fear about tomorrow," said Neil Chilson, head of AI policy at the Abundance Institute ^[13]. He pointed to the poll's finding that in every country but Spain, a third of respondents or more had rarely or never used AI ^[14]. "Most people haven't felt AI's benefits but they've heard many scary stories about its future," he added ^[15]. "For now, fear is outrunning experience. But experience will catch up" ^[16].

The results as reported do not break risk perception out by how often a respondent uses AI, so that prediction is untested here. Chilson could turn out to be right and the planning problem would still be real. A minister reading 45% to 51% pause support is reading the cost of defending a deployment in public ^[2], and that cost is set this year whether or not use converts fear into comfort later.

Outside the European Parliament buildings in Brussels on Monday, the PauseAI movement put it in blunter terms ^[17]. "We would not board a plane with a one-in-ten-chance of exploding mid-air," said Didier Coeurnelle, the coordinator of PauseAI Belgium ^[18], referring to an Anthropic employee's assessment of a higher than 10% chance that AI could wipe out humanity ^[19]. "That is what we are tolerating with AI today," he said ^[20].

Over the next two quarters, the 30% to 40% who want development to continue is the bloc a government counts before it says yes to anything contested ^[3]. The reported ranges make it the smaller of the two in every country measured ^[1]. For the decade, the question is whether experience catches up with fear ^[16]. This poll leaves that open. The lab principals have themselves raised slowing down, Dario Amodei, Sam Altman and Elon Musk all in recent days ^[21].