Musk proposes mutual pre-deployment reviews of frontier models
Speaking to Zanny Minton Beddoes, editor-in-chief of The Economist, in an interview published July 23, xAI founder Elon Musk called for rival AI labs to conduct peer reviews of each other's frontier models before deployment. The proposal, framed as a safety measure, was reported as Musk's response to what he described as an unusually compressed AI arms race, with multiple frontier systems already shipping.
Musk says founding OpenAI backfired
In the same interview, Musk said his original aim in helping create OpenAI was "essentially a counterweight to Google" when the search company "had more or less a monopoly on AI." He added that the subsequent spin-out of Anthropic, which he described as now leading in AI, had produced "knock-on effects that accelerated AI, which wasn't really my intention." Musk concluded: "It just seems like all roads lead to the acceleration of AI." He has long criticized OpenAI's evolution, according to the Business Insider report.
xAI's own disclosure record complicates the pitch
The peer-review proposal arrives against a credibility gap. Fortune reported in July 2025 that xAI released Grok 4 without a system card, the safety disclosure major labs use to publish model capabilities, limitations and risks. At the Seoul AI Safety Summit in May 2024, xAI joined Amazon, Anthropic, Google DeepMind, Meta, Microsoft and OpenAI in the Frontier AI Safety Commitments, which included public transparency around safety frameworks and risk management. The Startup Fortune analysis argued that a lab advocating pre-release disclosure cannot treat its own disclosure as optional.
Industry context: a compressed launch calendar
The proposal lands amid a tight cluster of recent product releases. OpenAI launched the GPT-5.6 family on July 9. xAI's developer release notes list Grok 4.5 on July 8, and the company's own launch page, dated July 16, described Grok 4.5 as its strongest coding and agentic work model. Anthropic restored Claude access from July 1 after export-control disruption, then confirmed wider top-tier subscription access later in the month, according to the report.
US government already runs a pre-deployment review track
A government version of Musk's idea already exists. The Commerce Department's Center for AI Standards and Innovation announced in May that Google DeepMind, Microsoft and xAI had agreed to give U.S. government scientists access to unreleased models for pre-deployment evaluation, joining OpenAI and Anthropic in voluntary reviews. The Guardian described the tests as focused on cybersecurity, biosecurity and chemical-weapons risks, with Reuters reporting that the expanded program gives U.S. scientists access to unreleased systems for risk assessments. The Startup Fortune analysis argued this structure has one advantage over Musk's sketch: the reviewer is not also trying to sell the rival product's replacement, though it would still need independent technical reviewers, strict confidentiality rules and clear risk thresholds to be workable.
Share this article
