The Pulse
EU to Convene Frontier AI Labs Over Pace of Advanced Models
European Commission President Ursula von der Leyen says she will invite leading frontier AI laboratories to discuss how to slow the development of self-improving models. She also calls for stronger cooperation with Canada and the United Kin
AI.info Team ·
European Commission President Ursula von der Leyen says she will bring the leading frontier AI laboratories together to discuss how to slow the development of increasingly capable, self-improving systems—a proposal that puts restraint alongside Europe’s demand to expand its own computing capacity.
“I will invite the main frontier labs for a discussion on how we can support ongoing industry efforts to pace the frontier,” von der Leyen said in her State of the Union address on September 16, 2026. She did not name the laboratories or set a date for the meeting.
Brussels wants slower advances and stronger safeguards
Von der Leyen framed the invitation as a response to risks associated with frontier models, including cyberattacks and systems that can improve their own capabilities. She cited concerns about AI agents escaping controlled environments or inserting malicious code, and said that executives at advanced AI companies are calling for a slower approach to self-recursive models.
“If the people developing the technology are clear, then we should be too,” she told lawmakers in Strasbourg. The Commission president said the EU’s AI Act would provide part of the framework, while international cooperation would address risks that cannot be handled by one jurisdiction alone.
Her proposed partners include Canada and the United Kingdom. Von der Leyen said the group should work together on model evaluation, verification, early warning systems and AI security, without specifying which agencies or companies would take part.
Europe’s restraint message collides with its race for capacity
The speech also makes clear that the EU does not intend to withdraw from advanced AI development. Von der Leyen says Europe must “stay in the race” to protect its national security and independence, and promises a major increase in computing capacity.
She says the Commission will present plans for expanding computing in a clean and efficient way. It also plans to develop new methods for combining public and private money to invest in European start-ups and other companies it considers promising.
That combination creates the central tension in the announcement: Brussels wants frontier developers to moderate the pace of the most autonomous systems while giving European companies more resources to compete. The speech offers no targets for new computing capacity, funding totals or limits on model development.
Von der Leyen shifts the focus from models to industry
Von der Leyen argues that Europe does not need to build every frontier model to gain economic value from AI. She says the region can draw on industrial data and established companies to develop specialised systems for real-world use.
The Commission is focusing on five sectors: health, transport, agri-food, advanced manufacturing, and defence and space. Von der Leyen says the EU will announce initiatives in those areas in November and invite member states, members of the European Parliament, industry and entrepreneurs to participate.
“The point is that AI needs to empower our doctors. Not to replace them,” she said while discussing AI-supported mammography and healthcare. Her broader position is that European AI should produce more value with lower cost and energy use, while preserving human control over consequential decisions.
The meeting’s unanswered questions
The planned discussion is a political invitation rather than a negotiated agreement. Von der Leyen does not identify which frontier labs will be asked to attend, whether the talks will produce voluntary commitments, or how any agreement would relate to the AI Act and existing safety work.
The speech also does not define what “pacing the frontier” would mean in practice. It could refer to slower training cycles, additional testing before deployment, limits on self-improving systems, or shared warning procedures. Those choices will determine whether the proposal becomes a forum for coordination or a wider effort to place boundaries around the next generation of models.
For now, the Commission has announced two linked aims: convene the companies building the most advanced systems and expand Europe’s ability to compute and deploy AI itself. The first concrete policy announcements are due in November, while the frontier-lab meeting has not been scheduled.