Demis Hassabis Calls for U.S.-Led Global AI Standards Framework

According to Google's AI division head, Demis Hassabis, the United States should take the lead in establishing rules to monitor emerging AI models. The organisation will conduct an additional evaluation of national security risks, which will encompass biological and cybersecurity threats.

Nobel laureate and Google DeepMind chief Demis Hassabis called for "urgent action" to mitigate dangers posed by artificial general intelligence (AGI) in a Tuesday piece published on X. This refers to the point at which AI achieves intellect levels comparable to or higher than humans. Frontier models have already shown to be a cybersecurity nightmare, he warned, and if technology develops further, new dangers, including nuclear and biological hazards, could materialise.

AI Needs Strict Monitoring: Hassabis

To combat the aforementioned dangers, Hassabis suggested a public-private partnership spearheaded by the United States and supervised by the federal government. Industry heavyweights have been demanding an AI watchdog for some time, but the public and private sectors can't seem to agree on how to regulate the most popular AI models. The Trump administration momentarily put export restrictions on a sophisticated model, leading Anthropic into weeks of negotiations with relevant authorities.

Additionally, OpenAI encountered limitations due to the fact that the US government had originally asked it to limit the release of a new model. According to Hassabis, the United States' economic and technological advantages put it in a prime position to spearhead the development of an AI framework. He elaborated by saying it might create a new standards body with a board that comprises independent top-tier technical specialists and open-source advocates, similar to the Financial Industry Regulatory Authority (FINRA) or a public-private partnership overseen by the federal government. In the United States, the exchange markets and broking businesses are overseen by FINRA.

How AI Watchdog will operate?

Hassabis stated that the proposed organisation would require "substantial" finance to recruit top-tier technical talent and supply the computing power required for extensive testing. Industry will "likely" provide the funding, he said. Once Frontier Labs' models are determined to be "effective", they will be required to voluntarily submit them to the agency for evaluation for 30 days prior to release in the United States.

According to Hassabis, certain agentic AI tests could detect attempts to evade safety measures and indications of dishonesty and guarantee optimal procedures. To help humans comprehend the thinking behind models, these methods should incorporate markers like digitally watermarking AI-generated images and creating tokens that can be read by humans as output. As the United States and China ramp up their competition to create and use AI models, there have been calls for stricter regulatory control.