The companies racing hardest to build more powerful artificial intelligence are suddenly finding common ground on a question that once divided them: is the race moving too fast?
On September 16, European Commission President Ursula von der Leyen backed efforts to slow the advance of frontier AI and said leading laboratories would be invited to talks on managing the risks.
According to Reuters, von der Leyen highlighted risks including the possibility that advanced models could be exploited for sophisticated cyberattacks and argued that governments and frontier labs need stronger risk-management mechanisms.
A rare agreement among fierce competitors
The striking part of the debate is not simply that regulators are worried. Some of the people building the technology are now publicly supporting additional safeguards.
The Associated Press reported that leaders associated with Anthropic, OpenAI and other major AI organizations have recently expressed support for slowing the pace of frontier development or strengthening oversight. Proposals include independent evaluators with deeper access to frontier laboratories, shared safety standards and greater international coordination.
OpenAI, Anthropic and Google have also been discussing AI safety directly, according to TechCrunch. That is notable because the same companies are competing intensely for models, customers, talent and computing infrastructure.
The problem: nobody wants to be the only company that slows down
Safety commitments become much harder when commercial and geopolitical incentives enter the picture. A company that deliberately delays a more capable model could fear losing customers or investment to a rival. Governments face a similar dilemma: tighter restrictions at home could be seen as handing an advantage to competitors abroad.
This creates a coordination problem. Participants can agree that uncontrolled development carries risks while simultaneously believing that slowing down alone would leave them worse off.
What would meaningful oversight look like?
Independent testing is emerging as one possible answer. Instead of asking the public to rely entirely on a laboratory’s own safety claims, external specialists could evaluate powerful models before or during deployment. Common standards could also make it harder for companies to gain an advantage simply by accepting more risk than competitors.
But major questions remain. Who chooses the evaluators? What information would companies have to disclose? Which capabilities would trigger additional restrictions? And how could international standards work when countries view advanced AI as both an economic asset and a national-security technology?
The AI debate is entering a different phase
For years, the central AI question was whether machines could become dramatically more capable. Increasingly, the debate is shifting toward what happens if they do.
The emerging agreement among rival AI leaders does not mean the industry has agreed on how much development should slow, which rules should be mandatory or who should enforce them. It does, however, show that concerns about frontier AI are no longer coming only from outside critics.
The people closest to the race are now debating the brakes. Whether those brakes become enforceable safeguards or remain voluntary promises made under competitive pressure may become one of the defining technology-policy questions of the next few years.
