Skip to news
PoliticsNews brief

EU Backs AI Slowdown and Invites Frontier Labs to Talks

Ursula von der Leyen says Europe will support a slower frontier race, putting recursive self-improvement at the center of global AI policy.

By THE COLDAI TIMES desk (AI-assisted)Editor: Shayan SalehiPublished 3 min read460 words

Key facts

  • Von der Leyen said she would invite frontier laboratories to discuss supporting efforts to slow the AI development race and proposed cooperation on evaluation, verification and early warning.[1][2]
  • The EU announcement came amid rare calls from leading AI executives for slower frontier development, while Meta’s Mark Zuckerberg rejected a coordinated slowdown.[2][3]
  • Von der Leyen’s position contrasts with President Donald Trump’s emphasis on maintaining U.S. AI competitiveness and skepticism about safety warnings.[2][3]
  • The proposal does not establish an immediate moratorium, leaving enforcement, thresholds and participation unresolved.[1][2]

The European Union has moved from observing the AI safety debate to trying to organize it. European Commission President Ursula von der Leyen said on Wednesday, September 16, that she will invite leading frontier AI laboratories to discuss how to support industry efforts to slow the development race around increasingly capable models.

The announcement came during von der Leyen’s annual State of the Union address in Strasbourg. She pointed specifically to models capable of recursive self-improvement, arguing that the risks are becoming more visible as systems gain the ability to produce code, operate agents and potentially contribute to the development of successor systems. The Commission president also proposed closer cooperation with countries including Canada and the United Kingdom on model evaluation, verification, early-warning systems and AI security. (france.representation.ec.europa.eu)

What changed

The EU’s move matters because it adds a major public regulator to a debate that had largely been driven by technology executives and researchers. Anthropic CEO Dario Amodei recently called for slowing frontier development, with support from OpenAI CEO Sam Altman and xAI chief Elon Musk. Meta CEO Mark Zuckerberg, however, has rejected the idea that companies need a coordinated industry-wide approach, saying each company should be responsible for making its own systems safe. (apnews.com)

Reuters reported that von der Leyen’s proposal puts her at odds with President Donald Trump, who has dismissed warnings about AI risks and emphasized maintaining the U.S. lead over China. The split highlights the emerging political fault line: whether advanced AI should be treated primarily as a strategic competition, an industrial opportunity or a technology requiring an internationally coordinated brake. (live.euronext.com)

Why it matters

The EU is not proposing an immediate moratorium in the announcement. Instead, it is seeking to turn voluntary industry concern into a broader framework for testing, monitoring and pacing advanced systems. That distinction could make talks more politically feasible, but it also leaves the central question unanswered: who would decide when a model has crossed the threshold for slower development, and what enforcement would follow?

Europe’s position is also deliberately dual-track. Von der Leyen said the EU must slow the race around self-improving models while massively expanding its own computing capacity and investing in European AI companies. In other words, Brussels wants stronger brakes on dangerous capability growth without surrendering technological independence.

That balancing act may define the next phase of AI policy. A shared evaluation and early-warning system could provide common evidence for governments and labs. But cooperation will be difficult while companies fear that restraint could hand rivals an advantage, particularly China. The EU’s invitation is therefore an opening move, not a settlement. The uncertain pieces are participation, scope and whether any eventual agreement would cover only frontier training runs or also deployment, access and open-weight releases.

Add The ColdAI Times as a preferred source on Google

Related stories