EU Invites Frontier AI Labs to Discuss Slowing the Race

The European Union is moving from AI rulemaking toward direct engagement with the companies building the most capable models. European Commission President Ursula von der Leyen said on September 16 that she will invite major frontier AI laboratories to discuss how Europe can support industry efforts to slow the pace of frontier development.

The announcement came in her annual State of the Union address to the European Parliament, where she framed increasingly capable AI as both an economic foundation and a security risk. She pointed to the possibility of models enabling more powerful hacking, as well as dangers from self-improving systems and autonomous agents that escape their operating environments or insert malicious code.

What changed

The proposal is not a new binding restriction, and the Commission did not identify a timetable, invitees or a specific definition of “slowing down.” Instead, it creates a diplomatic channel between European policymakers and frontier labs at a moment when industry leaders are publicly disagreeing over whether voluntary restraint is sufficient.

Von der Leyen also proposed cooperation with partners including Canada and the United Kingdom on model evaluation, verification, early-warning systems and AI security. That approach would extend Europe’s influence beyond the enforcement timetable of its AI Act, using technical coordination and international partnerships to shape how advanced systems are assessed before deployment.

Reuters reported that the invitation follows a broader split among technology executives and governments. Anthropic CEO Dario Amodei has urged companies to slow the rate at which they improve model capabilities, while OpenAI CEO Sam Altman and xAI’s Elon Musk have expressed agreement with the concern. Meta CEO Mark Zuckerberg, by contrast, has argued that competition, liability and individual company responsibility provide enough incentive for labs to act safely on their own.

Why it matters

The EU is attempting to occupy the middle ground between an open-ended moratorium and laissez-faire development. That matters because the most consequential risks described by policymakers—cyber misuse, autonomous behavior and recursive self-improvement—are difficult to address after a model has already been released widely.

The initiative also exposes Europe’s strategic tension. Von der Leyen said the bloc needs its own computing capacity and must remain in the AI race for economic and national-security reasons. Europe therefore wants stronger safeguards without surrendering access to the capabilities it believes will determine future productivity, defense and technological independence.

The practical test will be whether the talks produce measurable commitments: shared evaluation standards, incident reporting, pre-deployment testing or agreed thresholds for pausing capability increases. Without those mechanisms, the invitation risks becoming another high-level safety forum.

What remains uncertain is whether the labs will coordinate with one another, whether Washington will support Europe’s approach and how any voluntary commitments would interact with existing law. The next signal will be the Commission’s promised details on the meeting and whether November initiatives for industrial AI turn this safety agenda into concrete requirements across health, transport, agriculture, manufacturing, defense and space.