Technology and Science14 July, 20262 min read
Demis Hassabis Urges U.S.-Led Standards Body To Test Frontier AI Models Before Release
This article is an automated aggregation of coverage from news outlets. Our editorials offer in-depth analysis curated by the NewsCord team.
Hassabis urges a US-led independent standards body, FINRA-like, to test frontier AI before release. Independent regulator would test frontier AI before deployment, addressing national security risks.
Hassabis backs FINRA-style AI
Google DeepMind CEO Demis Hassabis called for “urgent action” and a U.S.-led oversight framework to test frontier AI models before they are released to the public. In an essay titled “A Framework for Frontier AI and the Dawning of a New Age,” Hassabis proposed a federally overseen public-private partnership or self-regulatory organization modeled on the Financial Industry Regulatory Authority (FINRA). The plan would create a Standards Body that sets and regularly updates benchmarks for “Frontier-class” models, with participating organizations deemed “Frontier Labs.”
Hassabis said Frontier Labs would voluntarily share models with the Standards Body for review up to 30 days before release, and that formal requirements could follow once the assessment protocol is shown to be effective. He also argued that the framework would support third-party auditors and testing in areas relevant to national security, while addressing risks that may surface as society moves further into the use of artificial general intelligence (AGI).
Cyber, bio, deception tests
Hassabis said the Standards Body’s testing could include “agentic AI tests” that look for attempts to bypass safety guardrails or signs of deception, alongside safeguards like digitally watermarking AI-generated images. He also described the need to “ensure best practices” such as generating human-readable output tokens to understand model reasoning, and he framed the work as a way to keep up with rapid capability change. CNBC reported that Hassabis proposed the body be overseen by the federal government and funded largely by industry to attract “world-class technical talent” and provide compute resources for large-scale testing.
In the same CNBC account, Hassabis said the U.S. is well positioned to lead because it has “economic and technical standing,” and he linked the proposal to national security risks including cybersecurity and biological threats. The proposal also comes as the U.S. and China race to develop and deploy AI models, with CNBC noting that U.S. lawmakers are considering how to curb adoption of Chinese AI models by homegrown companies.
Who decides, who pays
Hassabis’ proposal would shift decisions about release away from ad hoc government reviews, which TechCrunch said drew criticism for “lack of technical expertise and opaque decision-making.” TechCrunch reported that the standards body could be “backed by the U.S. government but funded by the AI industry and operated independently,” with staffing that includes open source representatives and technical experts. The debate over AI regulation also surfaced in TechCrunch’s account of White House AI advisor and a16z general partner Sriram Krishnan, who said “there will not be an FDA for AI.”
In parallel, CNBC said the White House, the State Department and the Department of Commerce have been approached for comment on Hassabis’ call for U.S. leadership. Across the coverage, Hassabis’ framework is presented as a way to make pre-release testing a prerequisite for deployment in the U.S. market once the assessment protocol is shown to be effective, while maintaining a pathway for ongoing post-release vulnerability work.