Anthropic, OpenAI and xAI spend most days trying to beat each other to the next AI breakthrough. Last week, the three companies agreed on something else instead: the industry needs to slow down.

Dario Amodei, Anthropic’s CEO, opened the conversation with an essay called We Must Pace the Frontier. His argument: AI systems are approaching the ability to help build the next generation of AI systems, a shift companies are pursuing without enough capacity to understand or control the result. Amodei pointed to a recent case of AI agents carrying out unauthorized cyberattacks as an early warning sign for more capable, less supervised systems. The extra time he wants, in his own framing, would go toward alignment research, better testing, and interpretability work, the effort to understand what happens inside a model while an answer takes shape.

His proposed plan has three parts. Independent evaluators would get employee-level access to check safety practices inside frontier labs. Frontier companies in democratic countries would agree to shared safety standards. And governments, including authoritarian ones, would work toward coordinated limits on how fast the technology moves. Anthropic committed to the first step on its own, promising outside evaluators permanent access to its systems.

Sam Altman answered within hours. “I agree with Dario that we need to pace the frontier,” he wrote, adding OpenAI had been discussing the same question internally for weeks and would share more soon. Elon Musk followed with two words: “Dario is right.”

OpenAI backed the words with an action. Altman ruled out a public listing for 2026, calling the timing ill-advised given safety concerns, and said the company still has “a lot of stuff to do” on how labs and governments coordinate on safety before revisiting the idea. He did not rule out a listing in future years. Some coverage of the moment has questioned whether spending pressure across the industry played a role too, alongside the stated safety concerns.

The essay followed a rough week for the industry. Days earlier, Jacob Coxon, a researcher who had worked at both OpenAI and Anthropic, resigned from the company, four months in and before his equity was due to vest, giving up the payout by leaving early. His resignation post, viewed more than 100 million times, warned: “The people building AI earnestly believe it could kill us all by the end of the decade. This is not a marketing stunt.” The exit carried particular weight because the lab has built its reputation on being the safety-focused option among its rivals.

Around the same time, Anthropic published its own account of hacking groups linked to China, Russia and Iran using Claude for espionage, propaganda and weapons-related research. One China-linked group alone was tracked targeting roughly 50 organizations worldwide, from government agencies to security vendors. The activity relied on stolen customer access credentials, and the company said its own systems were never compromised. The timing gave Amodei’s warning a concrete backdrop.

Part of the plan points directly at Beijing. Amodei wants frontier labs in democratic countries to agree on shared standards first, then use the resulting alignment to pull authoritarian governments, China above all, into the same limits. His essay argues a Chinese lead in advanced AI would be dangerous enough to justify chip export controls and limits on distillation, the practice of training a cheaper model by copying a more advanced rival’s outputs.

None of this is settled. Pacing the frontier remains a plan on paper. The plan depends on rivals who have spent years racing each other choosing to race more slowly instead. Whether safety explains the whole alignment or only part of the story remains unresolved in the coverage since Amodei’s essay. What is on the record is rarer than the reasoning behind the agreement: three companies competing for the same customers, the same talent and the same headlines told the public, on the same page, the industry should slow down.

Sources

Dario Amodei, We Must Pace the Frontier — darioamodei.com
SiliconANGLE, Sam Altman and Elon Musk back Dario Amodei’s call to slow down the frontier of AI development — siliconangle.com
Fortune, Sam Altman confirms OpenAI won’t go public this year, saying an IPO now would come at an ‘ill-advised moment’ given AI safety concerns — fortune.com
Axios, Scoop: Anthropic whistleblower gave up his equity to leave the company — axios.com
Anthropic, Countering Misuse of AI: September 2026 — anthropic.com