Analysis | 16 July 2026
When a Nobel laureate who runs the world’s most advanced AI research lab posts a 1,500-word essay, the internet pays attention. Demis Hassabis, CEO of Google DeepMind, does not publish often. When he does, it is usually worth reading.
His essay, titled A Framework for Frontier AI and the Dawning of a New Age, landed on X this week and immediately went viral. In it, Hassabis argues that humanity is standing at “the foothills of the singularity” and that the decisions we make in the coming months will determine the trajectory of civilisation for generations.
This is not abstract futurism. Hassabis has a concrete proposal, a specific timeline, and a growing coalition of industry leaders who agree with him.
The Argument: AGI Is Closer Than You Think
Hassabis believes artificial general intelligence – an AI system that matches or surpasses human cognitive ability – is “probably” only a few years away. DeepMind was founded in 2010 with 2030 as its horizon for AGI. That timeline has steadily compressed. Hassabis now puts AGI somewhere in the 2028-2031 window.
He compares the pending arrival of AGI to the discovery of fire and the invention of electricity. Those technologies fundamentally reshaped human civilisation. Hassabis argues AGI will do the same, but on a compressed timescale.
“The rapid progress we are seeing in AI requires a new approach to testing frontier AI model capabilities that is dynamic, adaptable, and rigorous,” he wrote.
The upside, he says, could be “unprecedented” – breakthroughs in medicine, energy, climate science, and every other domain where intelligence is the bottleneck. The downside, left unchecked, includes catastrophic risks from cybersecurity, bioweapons, automated deception, and loss of human control.
The Proposal: A FINRA for Frontier AI
This is where Hassabis moves from diagnosis to prescription. His proposal is specific enough to be actionable and practical enough to avoid the “FDA for AI” trap that has derailed previous regulatory efforts.
He wants a US-led, federally overseen Standards Body modelled on the Financial Industry Regulatory Authority (FINRA) – the private-industry watchdog that polices Wall Street. Not a government agency. Not an industry self-committee. A hybrid: funded by the AI industry, staffed by independent technical experts, and overseen by the federal government.
Here is how it would work:
Pre-release review. Frontier labs would voluntarily share new models with the Standards Body up to 30 days before release. Once the assessment protocol is proven effective, that review would become mandatory for deployment in the US market.
What gets tested. Hassabis wants mandatory evaluations for cybersecurity vulnerabilities, biological threats, deception capabilities, agentic risks, and national security implications. Specific tests would look for attempts to bypass safety guardrails, signs of strategic deception, and whether models can be trusted to operate autonomously.
Post-release oversight. Labs would work with the Standards Body to address any critical vulnerabilities discovered after deployment. This closes the window between release and responsible remediation.
International coordination. The framework is designed to expand globally, with independent auditors and the ability to coordinate a slowdown in frontier AI development if risks become too severe.
Funding. Hassabis is realistic about what this costs. “Funding would need to be substantial and likely mostly come from industry, in order to attract world-class technical talent and provide the necessary compute resources for large-scale testing,” he wrote.
Why Now? The Context Matters
Hassabis’s essay did not emerge from a vacuum. Several converging forces make this moment different from previous calls for AI regulation.
The US government recently imposed temporary export controls on Anthropic’s Mythos model and asked OpenAI to limit the rollout of Sol – the first times the federal government has directly intervened in a frontier model release. Those reviews drew significant criticism for lacking technical expertise and making opaque decisions.
White House AI advisor Sriram Krishnan recently told the Financial Times there “will not be an FDA for AI,” effectively ruling out a traditional regulatory agency. Hassabis’s FINRA model is a direct response to that constraint – a self-regulatory organisation that addresses the expertise gap without requiring a new federal bureaucracy.
The US-China AI race is also accelerating. Chinese models from DeepSeek and Z.ai are competing with leading Western frontier systems, and US lawmakers are actively considering how to curb the growing adoption of Chinese AI models by American companies. A US-led standards body could serve dual purposes: safety oversight and competitive positioning.
The Criticisms
No proposal of this scale is without its detractors. Critics on the industry side argue that pre-release review, even from an independent body, amounts to a licensing regime that will slow innovation and entrench incumbents. Smaller labs without the resources to navigate the approval process could be squeezed out.
Critics on the safety side argue that a FINRA model is still too close to the industry it is meant to regulate. FINRA itself has faced repeated criticism for being captured by the institutions it oversees. An AI Standards Body funded by the very labs it evaluates faces the same structural risk.
Chris Canal, CEO of AI safety evaluator Equistamp, told The Deep View that the body should “publish predictions of what the next generation of models will score before release, and be graded publicly on its accuracy. A testing regime that cannot predict is only measuring the past.”
The Bottom Line
Hassabis’s essay matters because it comes from someone who could build AGI. He is not a philosopher speculating about the future. He runs the lab that invented AlphaFold, AlphaGo, and Gemini. When he says we are at the foothills of the singularity, he means he can see the peak from where he stands.
The FINRA model is not perfect. But it is the most detailed, actionable proposal for frontier AI governance to come from within the industry. And it arrives at a moment when the US government is actively looking for something between doing nothing and creating an FDA for AI.
Whether the Standards Body materialises, and whether it can stay independent of the industry it regulates, will be one of the defining questions of the next phase of the AI age.
The most important decisions about AI are being made right now, by a handful of people, in a handful of rooms. Hassabis wants to change that.