Europe’s leading AI companies have pushed back against calls from US executives to slow the development of frontier models, accusing American rivals of using safety arguments to entrench their market position. The dispute, reported by Reuters on September 18, pits Mistral, Proton and Hugging Face against Anthropic’s Dario Amodei, who wants leading developers to coordinate voluntarily on how fast capabilities improve.
Amodei set out the position in an online essay, arguing that AI development is moving faster than governments can regulate. He proposed that leading developers voluntarily coordinate on safety standards and submit to oversight by permanent embedded evaluators, and called for an antitrust waiver that would let competing labs jointly develop safety measures without violating competition law. The essay followed warnings from researchers about the dangers of increasingly powerful systems, including scenarios in which advanced AI could pose existential risks to humanity.
European companies read the same proposals very differently. “Some incumbents are using this moment to consolidate their market position, pushing for regulation designed to favour them over competitors,” French startup Mistral said in a statement, noting that the risks Amodei cites had already been clear for months. Mistral, founded three years ago, is widely seen as Europe’s best shot at building a frontier model that can challenge US and Chinese labs.
Catch-up economics shape the debate
The disagreement tracks market position as much as philosophy. Europe has produced few frontier AI developers and relies heavily on models from OpenAI, Anthropic and Microsoft, leaving many on the continent wary of the ongoing dependence on US technology. A slowdown regime negotiated among incumbent US labs would freeze that hierarchy in place, which is exactly what European firms are trying to break out of.
Raphael Auphan, chief operating officer of Swiss privacy software company Proton, was blunt at a tech event in Paris. “I think this is totally self-serving,” he said. “They just want to preserve a dimension of dependency on their service.” Ben Brooks, head of public policy at German AI startup Black Forest Labs, argued that enforcing a gap between big labs and new entrants “on the basis of arbitrary thresholds will chill open innovation at or near the frontier.” Clement Delangue, CEO of open-source platform Hugging Face, put it shorter on X: “It’s not time to slow down but to accelerate.”
The split is not total. Notably, European critics did not reject every element of Amodei’s proposal. Independent evaluators embedded inside AI companies, the idea of permanent third-party oversight of frontier systems, drew support even from executives who rejected the slowdown framing.
Markets have already voted once
Investors took the safety-coordination debate seriously when it first surfaced. Reuters reported that AI-linked stocks slumped worldwide after top lab CEOs called for slowing the technology, a reminder that coordinated deceleration talk carries a price signal of its own. When the executives who know the technology best publicly argue it needs pacing, markets read it as information about risk, not just positioning.
OpenAI’s Sam Altman and xAI’s Elon Musk have both warned about risks from powerful AI systems and agreed on the need to pace development, which puts the US frontier labs broadly on one side of the argument and Europe’s scaling hopefuls on the other. The EU has its own regulatory track with the AI Act, but the current fight is about voluntary coordination among private labs, not statute, and the distinction matters because voluntary standards can move faster than law and reach further than any single jurisdiction’s rules.
What coordination would actually mean
Amodei’s concrete asks are two: an antitrust waiver so that competing labs can jointly develop safety standards without legal exposure, and permanent embedded evaluators monitoring compliance inside each company. He acknowledged the legislative path is too slow. “Unfortunately, passing laws can take time, and AI is advancing very quickly,” he wrote, per Reuters.
Critics see a problem with the mechanism regardless of intent. Voluntary coordination among the largest labs, even with a waiver, sets de facto industry standards that smaller players did not help write and may not be able to meet. Compute thresholds, evaluation regimes and safety certifications all cost money, and costs weigh heaviest on the firms with the least revenue. That is the competitive concern Mistral and others are voicing: safety architecture doubling as a moat.
There is also a geopolitical layer. China’s labs are not parties to any US-led slowdown pact, and European officials know that unilaterally pacing Western development while Chinese competitors continue at full speed would hand over strategic ground. That calculation pushes European governments toward supporting their own scaling efforts rather than endorsing restraint among US incumbents.
For now the dispute stays rhetorical. No binding coordination exists, the antitrust waiver is a request rather than a granted exemption, and the embedded-evaluator model is being tested through voluntary arrangements, including Anthropic’s engagement of outside assessment partners for its model releases. What has changed is that the safety-slowdown argument now has an organized, vocal opposition with names, quotes and market incentives attached, and that opposition is concentrated in exactly the region trying to build its own frontier capability.
The next test comes from deployment cycles rather than statements. If European labs ship competitive frontier models in the coming quarters, their argument that acceleration serves Europe’s interests gets stronger evidence behind it. If capability gaps widen instead, expect the safety-coordination camp to revive its case with regulators on both sides of the Atlantic, this time with the European startup community’s objections on the record.