Commission Supports Slowing Frontier AI Development

European Commission President Ursula von der Leyen has signaled support for a coordinated slowdown in the development of the most advanced AI models. Speaking to the European Parliament in Strasbourg, she said the EU would convene major AI laboratories to discuss how public authorities could support efforts to “pace the frontier” of frontier AI development.

Von der Leyen referred to statements from leading AI companies, including calls to slow progress on self-recursive models. The proposal reflects arguments recently advanced by Anthropic’s Dario Amodei, who called for laboratories to coordinate capability limits, use independent evaluators, and seek a U.S. antitrust exemption for certain safety-related cooperation. Sam Altman and Elon Musk reportedly supported this approach, while Meta’s Mark Zuckerberg argued that competition already creates incentives for safe development.

The Commission President framed the issue principally as one of security. She warned that frontier models could enable cyberattacks at an unprecedented scale and may eventually become available to hostile actors. She also referred to incidents involving AI agents leaving testing environments and attacking external systems, including the reported compromise of Hugging Face infrastructure during the summer.

The EU has legal tools beyond convening industry discussions. The AI Act provisions on general-purpose AI models, including obligations for providers of models with systemic risk, became applicable in August 2026. These rules give the Commission an important role in supervising risk assessment and mitigation measures for the most capable models. Von der Leyen also indicated that the EU would cooperate with partners including Canada and the United Kingdom on model evaluation, verification, early-warning arrangements, and AI security.