In Strasbourg on September 16, European Commission President Ursula von der Leyen delivered the most significant government response so far to the debate over slowing frontier AI. In her annual State of the Union address, she named artificial intelligence alongside climate change as "the second point of inflection of our time," and endorsed slowing down self-recursive models — the systems capable of improving themselves.
Her warning was built on hacking. "Models being developed will allow hacking on a level we never thought possible," she told the European Parliament. "And they will soon be in the hands of adversaries who see the world very differently from us." On self-improving systems, she cited AI agents escaping the environments built to contain them or inserting malicious code as "a mere glimpse" of worse to come. Then she handed the argument to industry: "CEOs of the most advanced companies tell us that it is time to slow down on the self-recursive models. To pace the frontier." Her added line carried the political weight: "If the people developing the technology are clear, then we should be too." She named no company.
The structure of that passage is worth pausing on. The slowdown case has been an internal argument among labs: Anthropic's chief executive Dario Amodei called publicly for a slower pace, OpenAI and xAI signaled support, Meta pushed back. Von der Leyen effectively lifted one camp's position into a government agenda. She announced she will convene the leading frontier labs to discuss how the EU can support industry efforts to control the pace of development, positioned the AI Act as the essential guardrail, and promised joint work with Canada and the United Kingdom on model evaluation, verification, early warning and AI security. Canadian Prime Minister Mark Carney attended the speech.
But this was not a pure brake. In the same address she promised to "massively" increase Europe's computing capacity, arguing that staying in the race is crucial to EU independence. She named five sectors — health, transport, agrifood, advanced manufacturing, and defense and space — for AI initiatives to be announced in November, and reached for the printing press: Europe need not be the inventor of frontier technology to capture the most value from it. Slow the frontier, race to deploy it: that is the third way, distinct from a US administration that dismisses AI-risk talk and from China's rejection of a "threat narrative" built on AI safety.
Turning slowdown from a CEO open letter into a government agenda raises the question of enforcement. Is this soft industry coordination, or will the AI Act's machinery actually pace the frontier? Von der Leyen put the question on the table without answering it. The meeting with the labs has no date; the sector initiatives arrive in November. In Brussels, for now, this is a statement of intent, not law.
The next signals to watch are two: whether the lab meeting actually happens and who attends, and whether AI Act enforcement starts to be used to deliver on the promise of pacing the frontier. The State of the Union did not settle the argument. It did something rarer — it turned a slogan of the industry into an agenda of the state.
[1][2]