The founder and CEO of Google DeepMind has called for the United States to establish clear criteria for identifying leading AI development organizations. In a recent blog post, he recommended forming an independent oversight group modeled on financial regulators to evaluate the most capable AI systems prior to their release.
The suggestion arrives as federal authorities examine cutting-edge models from several companies, sparking discussion on evaluation methods. The proposal includes a 30-day review period before public launch, positioned between current voluntary checks and longer formal processes.
In the post, the researcher stated that artificial general intelligence with human-level abilities may emerge within several years, requiring shared safety protocols. He described the present period as the early stages of a major technological shift comparable to the advent of electricity.
Such systems could accelerate progress in medicine, energy, and materials while boosting economic output at a scale exceeding the Industrial Revolution. At the same time, he noted that technical progress outpaces societal grasp of associated dangers, highlighting needs in cybersecurity, biological risks, and autonomous operations.
He advocated policies that encourage both innovation and accountability, along with global cooperation on safety measures. To implement oversight, he suggested a public-private standards organization that would set benchmarks, perform pre-release evaluations, and examine models for security threats or deceptive traits.
Developers would initially submit systems voluntarily, with possible mandatory requirements later. The framework would adapt over time through expert input and updated testing. While focused on the US, it could inform broader international guidelines for powerful AI.
