A Framework for Frontier AI and the Dawning of a New Age
This is a turning point in human history. Demis Hassabis believes artificial general intelligence (AGI)—a system with the full range of cognitive capabilities found in the human brain—may be only a few years away. Looking back decades from now, people may realize that this generation was standing in the foothills of the singularity, at the dawn of a new age.
AGI has been Hassabis’s life’s work. If built and deployed responsibly, it could become one of humanity’s most beneficial and transformative inventions. It belongs in a different category from the internet or mobile technology. It is closer to the discovery of electricity or fire. Humanity has, in effect, found a way to make sand think. That is astonishing.
The impact could be unprecedented: perhaps ten times the Industrial Revolution at ten times the speed. AGI could accelerate drug discovery, unlock new sources of clean energy, create advanced materials, and perhaps lead to an era when resources are no longer the main limit on human progress.
Mogu PSA:
A source check did not find a published method behind either “ten times” figure. Read them as Hassabis’s declaration of scale, not as numbers ready for a policy spreadsheet. The idea may be useful; the multipliers are not measurements. (¬‿¬)
The Challenges at the Frontier
AI is already producing real-world benefits, but realizing its larger promise requires navigating this critical development period carefully. Risks that may emerge closer to AGI need attention now. Frontier Models have already created cybersecurity challenges; nuclear and biological threats may follow as capabilities advance. Beyond that, humanity will need stronger safeguards to maintain control over more autonomous Agents, systems that can use one generation of results to improve the next, and problems that are still impossible to see clearly.
Human ingenuity can solve difficult problems, and society can manage AI’s technical risks—but only if it leaves enough time and room to get the next crucial step right. The field, and society more broadly, are not doing that today.
AI development is locked inside an intense, multilayered commercial and geopolitical competition. That pressure accelerates progress and its benefits, but frontier capabilities are advancing faster than human understanding. Nobody knows what happens next, and experts disagree. When uncertainty and stakes are both extreme, cautious optimism is the sensible strategy: public policy should encourage innovation while rewarding responsibility and security, support international cooperation on key safety problems, and make society think carefully about how AI is deployed for the public good.
A Standards Body for Frontier AI
Rapid AI progress requires a dynamic, adaptable, and rigorous way to test frontier capabilities. Given its economic and technical position, the United States could take the first step by creating a standards body built as a federally supervised public-private partnership or self-regulatory organization, similar to the Financial Industry Regulatory Authority (FINRA). Its board would include independent technical leaders and open-source representatives. The body would need substantial funding, probably mostly from industry, to attract world-class talent and pay for large-scale computing resources.
FINRA is not a US government agency. It is a private nonprofit self-regulatory organization funded by member fees and overseen by the Securities and Exchange Commission. Hassabis invokes it to combine industry speed with public supervision. The catch is obvious: the same industry may fund the body, be governed by it, and help shape its rules, so independence has to exist in practice, not only on paper.
The body would create assessment protocols and work with relevant federal agencies and US national laboratories on capabilities tied to national security. A model crossing designated thresholds on a regularly updated set of Benchmarks would qualify as Frontier-class; an organization that possesses such a model would become a Frontier Lab.
Frontier Labs would be encouraged to publish detailed model cards, maintain strong internal cybersecurity, vet key personnel, and devote sufficient resources to safety and security research.
At first, labs would voluntarily provide models for review up to thirty days before release. Once the assessment process proved effective and robust, it could quickly become formal: Frontier Models would have to pass before deployment in the US market. Labs would also work with the standards body on critical vulnerabilities discovered after release.
Assessments should use rigorous scientific evaluations across cybersecurity, biological threats, and other high-risk domains. Tests for agentic behavior could look for attempts to bypass safety safeguards or signs of deception. Deployment practices such as digital watermarking would also be checked. Models would produce human-readable output Tokens intended to help researchers understand their reasoning.
That final proposal needs an important caveat: readable reasoning is not necessarily a faithful account of what influenced a model’s answer. Anthropic’s research found that reasoning models often failed to mention hints that changed their conclusions. A readable trace can be one safety signal, but it cannot serve as a complete monitor of a model’s internal process.
The evaluations would be updated regularly, perhaps quarterly at first. Outdated or saturated tests would be retired and replaced. Early tests could be developed in consultation with Frontier Labs, but the standards body would eventually need enough technical capacity to create independent held-out tests that labs had never seen, reducing overfitting. Working with the US government, it could also foster an ecosystem of third-party auditors to run assessments and develop new ones.
The strength of this approach is its technical focus: it supports innovation while encouraging responsible behavior. It is designed to keep pace with an accelerating field and adjust as major risks become visible. If the situation became serious enough, requirements could be tightened, potentially including a coordinated slowdown among Frontier Labs.
Frontier status would carry prestige and would be open to any organization whose model met the assessment criteria. The framework could apply regardless of a model’s country of origin or whether it was open or closed. Smaller systems from startups or academia that remained below the frontier threshold would be exempt.
The public consequence depends on where the threshold and passing conditions are drawn. Too strict, and society may get models that are safe but barely useful. Too loose, and a lab that follows the rules may stop while a neighboring company—or another country—deploys genuinely dangerous capabilities. A standards body is not merely assigning labels; it is deciding who bears which risk.
A US-led effort would only be the beginning of shared international standards for frontier AI. The technology will affect the entire planet. Ideally, the framework would help countries reach consensus on managing the most serious risks while ensuring that everyone can access and benefit from AI’s opportunities.
The Future Is Not Yet Written
AGI could become the ultimate instrument for advancing science and medicine while driving enormous productivity and economic growth. Reaching that future requires sound technical foundations: coordination around a shared global framework, the most rigorous scientific methods available, and the best minds working together on the challenge.
Even if the technical problems are solved, harder economic and philosophical questions remain. What economic models would let everyone thrive in a world of abundance? What values should guide life? What will meaning and purpose become, and how might the human condition itself change? Those questions cannot and should not be left to technologists. Every part of society has to help define the next chapter.
AI inspires both enormous excitement and deep uncertainty, and both reactions are justified. The future is not yet written. Before AGI arrives, humanity still has a precious period in which to shape this technology for everyone’s benefit. The choices made now will determine how the next phase of civilization unfolds. If AGI can be brought safely into the world, it may open a new age of scientific discovery, progress, and human flourishing.