Co-founder and CEO Google DeepMind And Isomorphic LabsNobel laureate Demis Hassabis has called on the US to take an initiative to create a new organization to oversee artificial intelligence (AI). He stressed that the agency should have the right to test the world's most advanced models and coordinate the slowdown of their development throughout the industry if any danger arises.
In his Substack article, Hassabis noted that the United States could lead the development of a system of standards for AI. He proposed creating a body modeled on a public-private partnership under federal oversight, with the participation of independent technical experts and open source representatives.
As an example, he cited the Financial Industry Regulatory Authority (FINRA), which already exists in the United States, which operates on a similar model. He also mentioned the economic and technological position of the United States, thanks to which the country “has every opportunity” to take the first step in developing such a system.
“This US initiative will be a good starting point for creating common international standards in the field of advanced artificial intelligence. Ideally, this system will push the international community to reach consensus on how to manage the most serious risks,” Hassabis emphasized.
As the expert noted, advanced models create threats to cybersecurity. He also did not rule out the emergence of risks associated with biological and nuclear weapons as they develop. It is precisely to prevent such threats that a new organization should be created, Hassabis is sure.
According to the proposal of the CEO of Google DeepMind, the first rule is to establish the voluntary submission of AI models for evaluation to a new body 30 days before release, and after checking the effectiveness of such a mechanism, make the procedure mandatory for entering the United States market.
“The standards body will be responsible for developing evaluation protocols and collaborating with relevant U.S. federal agencies and national laboratories to conduct testing in areas relevant to national security,” Hassabis said.
In addition to technical problems, economic and philosophical issues must be resolved, the Nobel laureate believes. He noted the importance of building new economic models and understanding how the role of humans will change with the development of technology.
He paid special attention to the emergence of artificial general intelligence (AGI) – a system that has all the cognitive abilities of the human brain. According to Hassabis, this will be “a turning point in human history” that is likely “just a few years away.”
He himself compared the emergence of such a model to the singularity – a hypothetical moment in the future when the development of technology will reach such a level that it will become uncontrollable and inaccessible to human understanding.
In an interview with Axios, Hassabis added that the time has come for a more “systemic” approach to AI regulation. Today's cyber risks associated with artificial intelligence, according to the head of the Gemini project at Google, are “warning shots.” More serious dangers could emerge within 18 months, he said. The expert attributed this to the existence of open source models, which are not under the control of any government.
That's why Hassabis spent several months secretly drumming up support for a plan to create a single AI regulator, informing the administration of US President Donald Trump, colleagues from other laboratories and European officials before announcing the project publicly, the newspaper writes.
The Trump administration's impromptu crackdown on Anthropic's Mythos and Fable models last month was “something of a wake-up call,” Hassabis said, and proof that Washington needs something more substantive than ad hoc directives, Axios said.
Recall that the US government, due to security concerns, in June ordered the suspension of access to advanced AI models Claude Fable 5 and Mythos 5 for all foreign citizens located in the country and abroad, including foreign employees of the company Anthropic, which developed these neural networks.
