Governments can establish a system where only licensed AI companies gain access to large quantities of advanced chips, thereby controlling who can develop powerful AI models.

A collaborative effort among nations could lead to a global governance framework that oversees the entire lifecycle of AI models and their hardware requirements.

To harness the advantages of AI while managing its risks, nations must control the physical inputs to AI, paving the way for a secure and prosperous future.

In the U.S., proposed regulations require cloud computing providers to identify foreign customers who are training potentially dangerous AI models.

Reflecting a move towards governing AI technology, the U.S., Japan, and the Netherlands have implemented export controls on chip-making equipment to prevent sales to China.

Countries can adopt a regulatory framework for AI by controlling access to the specialized chips required for training advanced AI models.

Establishing such governance can prevent bad actors from accessing powerful AI while fostering innovation by allowing equitable access to computing resources.

The U.S. government is piloting national cloud computing resources to provide access to powerful AI models for academics and startups while mitigating risks of abuse.

An international collaborative effort similar to CERN for AI could enhance global scientific cooperation and safety in AI development.

The specialized chips necessary for advanced AI models are predominantly produced in Taiwan and rely on technology from Japan, the Netherlands, and the U.S., resulting in a tightly controlled supply chain.

Artificial intelligence (AI) holds the promise of significant societal benefits, but it also carries risks, including the potential for developing advanced weapons.

To govern AI effectively, it is essential to ensure that its benefits are widely shared while preventing dangerous AI from falling into the hands of malicious actors.