Musk Backs Call to Pace the AI Frontier, Citing Safety

SAN FRANCISCO — Elon Musk, whose xAI builds the Grok family of models, has thrown his weight behind an unusual idea gaining traction among the industry's top labs: that the race to build ever-more-powerful artificial intelligence should be paced deliberately, not run flat out. His endorsement puts three of AI's most competitive figures on the same page about safety.

Three Rivals, One Message

The catalyst was an essay titled "We Must Pace the Frontier," published over the weekend by Anthropic CEO Dario Amodei, who argued that model capabilities are now improving faster than researchers can understand and control them. "We must slow the pace at which we improve the capabilities of AI models," Amodei wrote, while stressing that he was not calling for a halt.

Musk's response was characteristically concise: "Dario is right." OpenAI's Sam Altman quickly added his own agreement, saying, "I agree with Dario that we need to pace the frontier," to give society time to adapt. That three leaders of fiercely competitive companies converged on the same position marks a genuine shift, as detailed in CoinDesk's report on the emerging consensus.

Long a Voice for Responsible AI

For Musk, the stance is consistent rather than surprising. He has warned about the risks of unchecked artificial intelligence for more than a decade, and he founded xAI in part to build what he described as a maximally truth-seeking and safety-conscious alternative. Backing Amodei's proposal reinforces Musk's long-held view that the technology is too consequential to develop recklessly, even as his own labs push the state of the art forward with Grok.

That balance — advancing capability while insisting on guardrails — has defined xAI's approach. The company has continued shipping fast, from Grok's integration into Tesla vehicles to deployments in sensitive settings, including Grok for government and defense work. Musk's endorsement suggests he sees no contradiction between building powerful systems and pacing the most dangerous capabilities.

What Pacing Would Look Like

Amodei laid out a three-stage framework that the rivals are beginning to embrace. It starts with independent evaluators embedded inside frontier labs with employee-like access, a step Anthropic and OpenAI both said they would adopt. From there, AI companies in democratic countries would coordinate on common safety standards, and eventually governments would negotiate international limits on the most dangerous capabilities, potentially including a "speed limit" on recursive self-improvement.

The concern driving the push is that AI systems are starting to automate the work of building more capable AI — a feedback loop that, left unchecked, could outrun human oversight. More than 1,300 employees across leading labs have signed on to a broader pacing initiative, signaling that the sentiment reaches well beyond the executive suite.

For an industry long defined by breakneck competition, the new alignment among Musk, Amodei and Altman is striking. It suggests the people building the most powerful technology of the age increasingly agree that winning the race should not mean moving as fast as technically possible — a message Musk has been sending for years, now echoed by his rivals.