European Commission President Ursula von der Leyen convened a group of senior officials this month for a closed-door meeting at a 19th-century palace on the outskirts of Brussels.
The meeting was intended to prepare a speech outlining the European Union's priorities for the coming year across its 27 member states, with artificial intelligence taking center stage.
The EU's new AI chief, Jim Hagemann Snabe, who serves as chairman of the supervisory board at German industrial giant Siemens, laid out how Europe could capture economic gains from AI.
According to two people familiar with the discussions, concerns about AI safety and mounting risks were also raised.
One person said that after the September 4 meeting concluded, von der Leyen and her commissioners have been grappling with how the EU should respond to a technology that is evolving at breakneck speed.
Should they tighten regulation to address emerging risks from AI? But if other countries do not impose equally strict controls, would tougher rules come with economic and geopolitical costs?
No final conclusions have been reached, and EU leaders will hold further discussions this year.
"AI offers both opportunities and risks," von der Leyen said in her September 16 State of the Union address. "If we cannot properly address these risks, we will never unlock AI's full potential."
The dilemma reflected in the Brussels meeting is playing out around the world.
Many governments are struggling to keep pace with a technology that is advancing far faster than their ability to respond.
AI evolves so rapidly that laws introduced just two years ago are already of limited effectiveness; numerous proposals to regulate AI have stalled and cannot reach a vote; and international summits on how to respond end with consensus quickly forgotten.
This newspaper conducted more than 20 interviews with lawmakers, tech workers and policy experts, and the findings show that AI is further widening the gap between policymaking and technological reality.
Many countries find themselves caught between wanting to fully capture AI's benefits and fearing its hidden risks.
Domestic politics, shortages of technical talent, geopolitical divisions and a reluctance to fall behind in a race that could determine economic strength and national security for decades are making the situation even more complicated.
"People always assume that on something this big, there will be mature, rational gatekeepers present," said Sean O hEigeartaigh, director of the AI Futures and Responsibility Project at Cambridge University and a member of the EU's AI advisory committee. "But the reality is that no such gatekeepers exist. That is precisely what makes the whole thing so deeply unsettling."
So a policy vacuum has formed, leaving AI companies such as Anthropic and OpenAI largely dependent on self-restraint.
Although the EU passed its AI Act in 2024, implementation will take years, and many argue that given the rapid pace of technological progress, the law already needs revision.
India, Japan, Australia, Canada and Kenya have taken a more lenient or wait-and-see approach; Brazil is discussing more comprehensive regulatory legislation.
Calls for government action are growing louder.
In particular, the heads of Anthropic, OpenAI and Google DeepMind have recently urged that development of the most advanced large models be slowed and government oversight strengthened.
Some researchers have modeled scenarios in which AI capable of self-improvement could escape human control and lead to catastrophic outcomes.
American public attitudes toward AI have also shifted, with recent polling showing that 61% of swing voters oppose building the data centers that support AI operations.
Comments