Google DeepMind CEO Demis Hassabis has proposed the creation of an independent, US-led standards body modeled after the Financial Industry Regulatory Authority (FINRA) to evaluate and regulate frontier AI models before public release, warning that humanity has a "precious window" to ensure artificial general intelligence is developed safely.
In a framework document titled "A Framework for Frontier AI and the Dawning of a New Age," published on X on Tuesday morning, Hassabis laid out a detailed proposal for how the United States could establish a credible regulatory infrastructure for the most powerful AI systems. The plan calls for a new organization that would test frontier models and develop best practices for their deployment — funded by the AI industry but operated independently of any single company.
For the latest breaking AI news and in-depth policy analysis, AI Buzz Wire tracks the developments shaping the industry.
How the Proposed Standards Body Would Work
Under Hassabis's proposal, AI laboratories would voluntarily share their frontier models with the standards body for review up to 30 days before public release. This pre-deployment assessment would evaluate models for safety, security, and potential misuse risks. Once the assessment protocol is proven effective and robust, the framework envisions formalizing the process so that frontier models would be required to pass it before being deployed in the US market.
The proposal directly addresses a gap exposed by recent government reviews of frontier AI systems. Earlier this year, the US government conducted ad hoc safety reviews of Anthropic's Mythos model and OpenAI's Sol model before their respective releases. Those reviews drew significant criticism for lacking technical depth and for opaque decision-making about when a model was deemed safe enough for public deployment.
Under Hassabis's framework, those decisions would be handled by a dedicated organization staffed with technical experts, backed by the US government but insulated from political pressure. Labs would also be required to work with the standards body to address critical vulnerabilities discovered after release.
A 'Precious Window' Before AGI
In interviews with Axios, CNBC, and The Economist accompanying the framework's release, Hassabis emphasized that the current moment represents a narrow opportunity to establish governance mechanisms before artificial general intelligence arrives. He described this period as a "precious window" — implying that the window will not remain open indefinitely.
Hassabis has long been an advocate for proactive AI governance. His framework arrives amid intensifying debate over whether the United States should create a formal AI regulator, and if so, what form it should take. The FINRA model is notable because the financial industry's self-regulatory organization is funded by member firms but operates under government oversight, giving it both industry expertise and public accountability.
Resistance from the Trump Administration
The proposal faces significant political headwinds. The Trump Administration has signaled deep skepticism toward creating new regulatory bodies for artificial intelligence. White House AI advisor and a16z general partner Sriram Krishnan — who has since departed his role — publicly dismissed the idea of an AI regulator within the executive branch, stating bluntly that "there will not be an FDA for AI."
That stance has placed administration officials at odds with some of the industry's most prominent leaders. While companies like Anthropic and OpenAI have broadly supported pre-deployment safety testing, the mechanism and authority for such testing remain contested. Hassabis's proposal attempts to thread this needle by positioning the standards body as industry-funded and voluntary at first, with formalization coming only after the process proves its value.
Building on Existing Efforts
The framework does not start from scratch. It builds on the voluntary commitments that several frontier AI companies have already made regarding safety testing and red-teaming. It also draws on international efforts, including the UK's AI Safety Institute and the G7's Hiroshima AI Process, which have established preliminary frameworks for evaluating advanced models.
However, Hassabis's proposal goes further by calling for a permanent, well-resourced organization with a mandate to test every frontier model before release — a level of scrutiny that no current body provides. The FINRA analogy suggests a scaled, professional operation with hundreds of technical experts, not a small advisory panel.
The proposal also addresses post-deployment monitoring, requiring labs to work with the standards body on vulnerabilities discovered after a model is publicly available. This reflects growing recognition that safety risks do not end at launch, as demonstrated by jailbreaks and misuse cases that have emerged after several high-profile model releases.
Industry Response and Next Steps
The framework has generated significant attention across the AI policy community. Supporters argue that a credible pre-deployment testing regime could build public trust in frontier models and reduce the risk of catastrophic misuse. Critics, including some free-market advocates, warn that a mandatory testing requirement could entrench large incumbents by creating regulatory barriers for smaller companies and open-source developers.
Hassabis's proposal does not specify which models would qualify as "frontier" — a threshold question that will determine the scope of any future regulatory regime. It also leaves open how the standards body would handle classified or dual-use applications, particularly given recent concerns about AI models being used to crack classified US systems.
The timing of the proposal is deliberate. With AGI appearing increasingly plausible within the next several years, Hassabis is pressing the case that the governance infrastructure must be built now, not after a dangerous capability emerges. Whether Washington will act on his call remains to be seen.
Stay Ahead of AI
For comprehensive coverage of AI policy, regulation, and industry developments, visit AI Buzz Wire.
Read more AI news →
