Elon Musk is calling for the world's most advanced AI models to undergo peer review by competing laboratories before public release—but he wants the industry, not the government, to take the lead.

The Tesla and SpaceX CEO said leading AI companies should collaborate to self-police and flag safety concerns tied to the cutting-edge models they are deploying at an accelerating pace. Musk has been among the most prominent voices raising alarms about AI risks for more than a decade. In 2023, he was a signatory to an open letter from the Future of Life Institute that called for a six-month pause on training AI systems more powerful than GPT-4—a request the industry did not heed.

"The most immediate thing that we could do is to have the leading AI companies at least just meet—have some sort of call—once every few weeks and just discuss any safety and security issues," Musk said in an interview with The Economist editor-in-chief Zanny Minton Beddoes published Thursday.

Musk argued that the government should intervene only if a company identified a security problem it was unwilling to fix. Otherwise, granting rivals a week or two of early access to review new models would allow "competitors can keep each other honest," he said.

"I think it's quite difficult for someone in the government who doesn't have a deep technical understanding and isn't driving the frontier of AI to know whether something should be released or not," Musk said.

When Minton Beddoes asked when such a system should be put in place, Musk replied, "immediately," adding that even a six-month delay would be excessive given the speed of AI development.

"There's so many AI breakthrough announcements, sometimes multiple per day," he said. "I mean, how many were there last week?"

OpenAI, Anthropic, Google, and Musk's own xAI—now part of SpaceX—have all released new AI models within the past month as competition among leading labs intensifies.

Musk's remarks follow an OpenAI disclosure earlier this week that two of its AI models escaped from a "sandbox," an isolated testing environment with no direct internet access, and infiltrated the systems of Hugging Face, an open-source platform for hosting AI models. According to a company blog post, the models acted autonomously to cheat on an internal evaluation test.

The incident heightened concerns about the growing capabilities of AI agents and their potential to autonomously execute sophisticated cyberattacks. Although Musk's interview was recorded Monday—before OpenAI publicly disclosed the breach—he wrote in a Wednesday post on X: "We are in the Singularity." The comment, made in response to another user's post summarizing several recent AI advancements including the Hugging Face incident, appeared to reference the hypothetical moment when AI surpasses human intelligence and drives technological change at a pace humans can no longer anticipate or manage.

Experts have previously told Fortune that advanced AI models in the hands of malicious actors pose a significant threat to organizations. That includes Anthropic's Mythos model, which the company initially withheld from public release because of its remarkable ability to uncover cybersecurity vulnerabilities, as previously reported.

AI enables malicious actors to carry out cyberattacks even with minimal coding knowledge, said Zach Lewis, chief information officer and chief information security officer at the University of Health Sciences and Pharmacy in St. Louis. AI agents could further lower the barrier to entry by automating the attack process, Lewis noted.

In June, President Donald Trump signed an executive order asking AI companies to voluntarily grant the government oversight of unreleased AI models. The order built on a broader set of voluntary safety commitments that several major AI companies made at the White House in 2023. Following the Hugging Face incident, some lawmakers, including Rep. Greg Casar (D-Texas), have called for stricter AI regulation.

Thomas Wolf, cofounder of Hugging Face, told the BBC that the attack on his company is a "wake-up call" for the industry and warned that "this will be one of the most common types of cyberattacks we see" going forward.

Musk's proposal for inter-laboratory collaboration would build upon the existing Frontier Model Forum, an industry body established in 2023 through which Amazon, Anthropic, Google, Meta, Microsoft, and OpenAI voluntarily share information about vulnerabilities and threats. Musk's xAI is not a member.

However, the companies in the Frontier Model Forum do not share unreleased models with one another—a gap that Musk's proposal would directly address.

Some firms, including OpenAI, have already provided the government with early access to certain models. In a blog post from late June, the company stated: "As part of our ongoing engagement with the U.S. government, we previewed our plans and the models' capabilities," ahead of the launch of its latest model, GPT-5.6 Sol.

OpenAI noted in the post that it did not expect government pre-release sharing to become standard practice, but rather described it as "taking this short-term step because we believe it is the strongest path to broader availability in the coming weeks."

Musk acknowledged that meaningful cooperation among leading AI companies would require him and OpenAI CEO Sam Altman to set aside their differences. The two have been locked in an escalating dispute that traces back to Musk's role as a co-founder of OpenAI in 2015 and his departure from the organization three years later. They have recently traded barbs on social media and have been clashing in court over the SpaceX CEO's opposition to OpenAI's transition from a nonprofit to a for-profit entity.

"At the end of the day, if we have to talk, we'll talk. I mean, set aside our personal differences for the good of the world type of thing," Musk said.