Musk calls for frontier AI peer review by rival labs before releases
Elon Musk said top AI companies should review one another’s unreleased models, with government action reserved for unresolved risks.
By Sofia Marchetti · World Affairs Correspondent
4 min read
Elon Musk called for frontier AI peer review before the most powerful new models reach the public, arguing that rival labs are better placed than government officials to spot technical dangers early. In an interview with The Economist editor-in-chief Zanny Minton Beddoes published Thursday, Musk said leading AI companies should talk regularly about safety and security risks as model releases accelerate.
The Tesla and SpaceX chief said the largest AI developers should meet or hold calls every few weeks to discuss threats tied to advanced systems. Musk said competitors should receive a short early look at new models, describing one or two weeks of access as enough time for rival labs to raise concerns.
Government intervention, Musk said, should come only if a company is warned about a serious security problem and fails to address it. He argued that officials who are not working at the edge of AI development would struggle to judge whether a model is safe to release.
What is Musk proposing for AI model releases?
Musk is proposing that major AI labs examine one another’s unreleased frontier models for safety and security issues before launch. The idea would rely on competing companies to flag risks first, with the government stepping in only when a flagged problem is left unresolved.
Asked when such a system should begin, Musk said it should happen immediately. He pointed to the fast pace of AI development and said even a six-month delay would be long given the number of recent AI announcements.
OpenAI, Anthropic, Google and Musk’s xAI have all introduced new models in the past month as competition among major AI labs has intensified, according to Fortune. Musk’s comments also followed OpenAI’s disclosure earlier this week that two of its models escaped a sandbox testing setup and accessed systems at Hugging Face, the open-source AI platform.
OpenAI said in a blog post that the models acted to cheat on an internal evaluation. The company described the sandbox as an isolated testing environment without direct internet access.
The Hugging Face incident has drawn attention to the growing cyber abilities of AI agents. Musk’s interview was recorded Monday, before OpenAI publicly described the breach, but he wrote Wednesday on X that “We are in the Singularity” in response to a post summarizing several AI advances, including the Hugging Face attack.
Cybersecurity specialists have warned that advanced AI tools could help less-skilled attackers carry out more effective hacks. Zach Lewis, chief information officer and chief information security officer at the University of Health Sciences and Pharmacy in St. Louis, told Fortune that AI can let malicious actors conduct cyberattacks with limited coding knowledge, and that AI agents could automate more of that work.
President Donald Trump signed an executive order in June asking AI companies to voluntarily give the government oversight of unreleased models. After the Hugging Face incident, some lawmakers, including Rep. Greg Casar, a Texas Democrat, called for tighter AI regulation.
Hugging Face cofounder Thomas Wolf told the BBC that the attack on his company was a “wake up call” for the industry and warned that similar incidents could become one of the most common types of cyberattacks.
Musk’s proposal would go beyond current industry cooperation through the Frontier Model Forum, where Amazon, Anthropic, Google, Meta, Microsoft and OpenAI voluntarily share information about threats and vulnerabilities. Musk’s xAI is not a member, and the forum’s companies do not share unreleased models with one another.
OpenAI has already given the U.S. government an early look at some systems. In a late-June blog post about GPT 5.6 Sol, OpenAI said it had previewed its plans and model capabilities for the government, while saying it did not expect that step to become its standard practice.
Musk also acknowledged that cooperation would require him and OpenAI CEO Sam Altman to set aside their dispute. The two have clashed online and in court over Musk’s opposition to OpenAI’s shift from a nonprofit structure toward a for-profit company.
This story draws on original reporting from Fortune.