The heads of some of the world’s leading artificial intelligence companies are showing an unusual degree of agreement over one of the industry’s most difficult questions: how quickly should frontier AI systems be developed?

The debate gained momentum after Anthropic CEO Dario Amodei said he had “become convinced” that AI companies need to pace the development of increasingly powerful systems to reduce the risk of a potentially catastrophic outcome.

Within hours, several prominent figures across the AI industry publicly backed the proposal, including OpenAI CEO Sam Altman and Elon Musk, whose rivalry with Altman has often put the two on opposing sides of major technology debates.

“I agree with Dario that we need to pace the frontier,” Altman wrote on X.

Altman also said OpenAI would commit to a measure previously proposed by Anthropic: allowing independent, third-party safety evaluators to work inside AI companies with access comparable to that available to employees.

The proposal represents a significant shift from relying primarily on companies’ internal safety teams to assess the risks associated with increasingly capable models. Giving external evaluators employee-level access could allow independent researchers to examine systems more closely and potentially identify dangerous capabilities or failures that internal teams might overlook.

Musk also endorsed Amodei’s position, writing on X: “Dario is right.”

The support is notable given Musk’s long-running disagreements with Altman and OpenAI. Musk was among the founders of OpenAI but later left the company and has since launched his own AI venture, xAI, while repeatedly criticizing OpenAI’s direction and governance.

The unusual convergence among competing AI leaders is seen as an indication that concerns about the speed of frontier AI development are becoming harder for companies to treat as a purely internal matter.

Hugging Face CEO Clement Delangue also joined the discussion, saying his company was launching an “open alignment initiative” and asking to be included among AI labs that employ third-party safety evaluators.

Andrej Karpathy, an AI researcher and former OpenAI and Tesla executive, similarly endorsed Amodei’s proposal for closer monitoring of AI development.

“I love this and really hope we can come together as an industry and make it happen,” Karpathy said on X.

From Competition to Collective AI Safety

The public support comes at a time when AI companies are competing aggressively to build systems with greater reasoning, autonomy, and scientific capabilities. That race has produced rapid advances, but it has also intensified questions about whether safety research and oversight can keep pace with model development.

Amodei’s argument has stirred a lot of interest because it does not amount to a call to stop AI development altogether. Instead, the proposal focuses on managing the rate at which frontier capabilities advance, creating more time for safety testing, evaluation, and governance to catch up.

That matters for an industry whose commercial incentives favor being first to market with increasingly capable models. A company that slows development unilaterally could worry about losing ground to competitors that continue moving rapidly.

The prospect of several major laboratories adopting common safety practices could therefore address one of the central problems in AI governance: the difficulty of asking individual companies to sacrifice speed when their rivals have no obligation to do the same. Independent evaluators could also provide a degree of external scrutiny at a time when AI companies are increasingly responsible for judging the safety of systems they have enormous commercial incentives to advance.

Google DeepMind founder Demis Hassabis had not publicly responded to Amodei’s Saturday proposal at the time, although he has previously argued for greater cooperation between AI companies and governments.

In a lengthy post on X in July, Hassabis called for public policy that supports innovation while encouraging responsibility and security. He also argued for international cooperation on important AI safety issues and greater consideration of how AI systems are deployed for society’s benefit.

That position points to the broader challenge facing the industry. Even if major AI laboratories reach agreement on voluntary safety measures, the development of powerful AI systems is unlikely to remain solely a corporate issue.

The technology is already spreading across workplaces, governments and consumer services, while frontier models are becoming more capable of carrying out complex tasks with limited human intervention. The consequences of failures, misuse or unexpected capabilities could therefore extend well beyond the companies building the systems.

The sudden support for Amodei’s proposal from rivals who normally compete fiercely over AI talent, products and market share may indicate that the industry is beginning to recognize a common problem: the race to build more powerful AI creates incentives to move faster, while the consequences of moving too quickly could eventually be shared by everyone.

The harder issue is public agreement translating into enforceable practices. Third-party evaluators, independent access, and coordinated safety standards could provide stronger oversight, but only if companies are willing to accept scrutiny that may expose weaknesses in their systems or slow commercial development.