# Elon Musk wants rival AI labs to vet each other's models before release > Source: > Published: 2026-07-24 00:04:23+00:00 *In a wide-ranging Economist interview published July 23, Musk proposed mandatory peer review of frontier AI models by competing labs, positioning rivals as better safety arbiters than governments. The proposal arrives as xAI, OpenAI, and Anthropic all launched major models in the same month.* The pitch sounds reasonable on its face. Leading AI companies should hold regular calls, Musk told The Economist in an interview taped on July 20, to "discuss any safety and security issues" and give competitors time to assess major new models before they go live. If a company ignored serious concerns flagged by its peers, "that would be the moment for government to step in and take action." Rival developers, he argued, are simply better placed than regulators to spot technical risks in systems that governments don't yet fully understand. That's the case, at least in theory. Here's the thing: the interview landed in the middle of an unusually compressed AI arms race. Grok 4.5, GPT-5.6, and Anthropic's Fable 5 all reached the market in roughly the same window in July 2026. Musk referred to Grok 4.5 as "Opus-class" in the interview - a capability claim that put it in direct competition with the systems he was simultaneously proposing should be reviewed. The interview was recorded before OpenAI disclosed that one of its frontier models had exhibited troubling autonomous behaviour during testing, a detail that would have made the peer-review argument considerably easier to make. Any governance proposal from a competitor in an active race deserves scrutiny on motive. This one doesn't hide its pressure points. A mandatory review period before deployment would impose the same delay on every lab, but the lab that is slightly behind on a given model cycle benefits most from slowing everyone else down. xAI is well-funded and technically capable, but OpenAI and Anthropic have both shipped models in rapid succession this summer. A structured pause, even a brief one, is more valuable to whoever is catching up. The credibility problem runs deeper than strategy. As Fortune reported, xAI released Grok 4 without a system card - the industry-standard safety disclosure that Anthropic and OpenAI both publish alongside major launches. At the Seoul AI Safety Summit in May 2024, xAI, alongside every other major frontier lab, signed the Frontier AI Safety Commitments, which explicitly included transparency around model capabilities, risk assessments, and known failure modes. Proposing that competitors submit to peer review while skipping your own safety report is not a coherent safety policy. It's positioning. Musk has been calling AI more dangerous than nuclear weapons since at least 2014. That's a long-standing view, and it's not obviously wrong. But the gap between the rhetoric and xAI's actual disclosure practices is wide enough that the peer-review proposal reads less as a principled safety framework and more as a governance shape that happens to suit xAI's current competitive position. ## What a real framework would actually require Strip out the self-interest question, and there's a legitimate idea buried in the proposal. The Commerce Department's Center for AI Standards and Innovation has already moved in roughly this direction: Google DeepMind, Microsoft, and xAI have agreed to provide access to unreleased models for pre-deployment security and capability evaluations, joining Anthropic and OpenAI in voluntary government reviews. Every major U.S. frontier lab now participates in some form of pre-release assessment. The infrastructure, at least in embryonic form, exists. The harder problem is that peer review among direct competitors is structurally different from peer review in academic science. In academic science, the reviewer has no commercial stake in whether the paper ships. In AI, the reviewer is also trying to win the same market. Asking OpenAI to flag risks in a Grok model, or xAI to assess GPT-5.6, creates incentive structures that are almost impossible to separate from competitive intelligence gathering. You'd need genuinely independent technical reviewers with enforceable confidentiality agreements and no equity stakes in any of the labs - which describes almost no one with the relevant expertise right now. None of that means the idea is worthless. It means the version Musk sketched in The Economist - companies holding calls every few weeks - is a starting sketch, not a workable framework. Translating it into something with real teeth would require agreed definitions of what constitutes a "frontier" model, binding timelines, clear criteria for what counts as a serious concern, and an enforcement mechanism beyond "then government steps in." None of those details were in the interview. For founders and developers building on frontier models, the governance direction matters regardless of how this specific proposal lands. Any formal pre-deployment review regime, even a voluntary one, changes release timelines and creates new liability questions around what a lab knew before shipping. The Commerce Department's existing voluntary framework could harden into something more binding quickly if a high-profile incident accelerates political pressure. Musk's proposal may not become policy. But the category of policy it describes is coming either way - and the current window, while the frameworks are still being drafted, is the right moment to pay attention. **Also read:** [The five biggest cloud builders are spending every dollar they earn on AI and then some](https://startupfortune.com/the-five-biggest-cloud-builders-are-spending-every-dollar-they-earn-on-ai-and-then-some/) • [New York becomes the first US state to ban new hyperscale data centers and other states are watching](https://startupfortune.com/new-york-becomes-the-first-us-state-to-ban-new-hyperscale-data-centers-and-other-states-are-watching/) • [Amazon brings Alexa Plus to UK browsers and the AI assistant wars get harder to ignore](https://startupfortune.com/amazon-brings-alexa-plus-to-uk-browsers-and-the-ai-assistant-wars-get-harder-to-ignore/)