Hassabis's AI Safety Plan
Google DeepMind CEO Demis Hassabis proposes US-led global AI watchdog for frontier model safety
Anonymous
Key Points
1
Frontier AI labs would voluntarily share their models with the new watchdog up to 30 days before release for safety testing.
2
The governing body would include representatives from open source communities alongside top independent experts, broadening participation beyond traditional labs.
3
Hassabis is described as a Nobel laureate behind the Gemini AI project, underscoring his credibility in the field.
4
Hassabis describes discussions with the administration as positive, noting he operates from a London base while pursuing U.S.-led governance.
5
The manifesto outlining the plan is titled 'A Framework for Frontier AI and the Dawning of a New Age,' articulating the proposed framework.
Google DeepMind CEO Demis Hassabis is calling on the United States to lead a new global AI watchdog — a body that could force top AI labs to submit their most powerful models for safety testing before release. According to CNBC, Hassabis wants the organization running within months, potentially before the end of this year.
The proposal comes as Hassabis warns the window to lock in AI safety is closing fast. He says advanced AI capabilities could leak into open-source models within 18 months, raising risks to cyber, biological, and nuclear security. Cryptopolitan reports he has already spent months briefing the Trump administration and European officials to build support.
A FINRA-Style Body to Screen Frontier AI Models
Hassabis wants the new watchdog modeled on FINRA, the self-regulatory body that oversees U.S. financial brokers. According to FourWeekMBA, the proposed body would be privately funded but answerable to the U.S. government. It would be staffed by top independent technical experts and would define which companies qualify as "frontier labs."
Under the plan, frontier labs would voluntarily submit their AI models up to 30 days before public release. The watchdog would then run tests for cyber threats, biological risks, and deception. The body would also include voices from open-source communities, not just big labs. CNBC notes the goal is to screen systems before they reach the public, not after.
Hassabis Warns the Safety Window Is Shrinking Fast
Hassabis laid out his thinking in a manifesto titled "A Framework for Frontier AI and the Dawning of a New Age." He argues the next phase of AI development is unlike anything before it. He says capabilities are advancing so quickly that safety standards must be set now — before the technology outpaces any oversight.
His biggest fear is the 18-month timeline. He believes cutting-edge AI capabilities could soon appear in open-source models that anyone can download and use freely. At that point, controlling how those models are used becomes nearly impossible. Cryptopolitan reports he sees risks not just to tech but to nuclear and biological security on a global scale.
Nobel Laureate Lobbies Trump White House and European Leaders
Hassabis is a Nobel laureate — recognized for his work on protein structure — and the architect of Google's Gemini AI project. According to GuruFocus, that credibility has helped him get serious hearings in Washington. He describes his talks with the Trump administration as "positive," even while operating from his base in London.
He has also briefed European officials, aiming to build a broad coalition behind U.S. leadership on AI governance. TechBuzz reports that many leaders at major AI labs have signaled they agree with the goal, even if details remain unsettled. The U.S. role is key: Hassabis believes American economic and geopolitical leverage is what makes a global standard enforceable.
Critics Fear Power Concentration and Slower Innovation
Not everyone is on board. FourWeekMBA points out that a FINRA-style body would also control who gets labeled a "frontier lab" — and who does not. That power could lock out smaller players and startups while giving established giants like Google DeepMind a built-in structural advantage.
Critics also warn the 30-day review window could slow the pace of AI releases and put U.S. labs at a disadvantage against foreign competitors who face no such rules. Supporters counter that the alternative — no standards at all — is far more dangerous. The debate now shifts to whether Washington will act before the 18-month clock runs out.
Sentiment
Mostly NegativeNEGATIVENEUTRALPOSITIVE
Publishers
20
Articles
6
Reach
26
A
AnonymousAuthor