Google DeepMind CEO Demis Hassabis shared a post outlining a detailed proposal on how to regulate top-tier AI models. He believes that true Artificial General Intelligence (AGI) could emerge within just a few years, bringing changes that are 10 times bigger and 10 times faster than the Industrial Revolution. However, the current pace of AI development has already surpassed humanity's ability to fully understand it. He suggests that the U.S. should take the lead and establish a dedicated 'AI Standards Agency,' modeled after regulatory bodies in the financial industry, with members including independent technical experts. The agency's role would be to test models before their release—focusing on their capabilities in high-risk areas like cybersecurity and biosafety. If a model reaches a certain danger level, it would be classified as a 'frontier model' requiring heightened regulation. Initially, major AI companies could voluntarily submit their models to the agency for testing up to 30 days before official release. Once this testing mechanism proves effective, it could become mandatory in the future—meaning any top-tier AI model launched in the U.S. would need to pass this checkpoint first. These rules would apply regardless of the model's country of origin or whether it’s open-source, but smaller startups or models developed by schools would generally be exempt.