Google DeepMind CEO proposes independent body to oversee AI

Google DeepMind CEO Demis Hassabis has proposed the creation of a new international standards body to regulate the most advanced AI models and ensure their safety. This proposal is causing a stir in the tech world, as it aims not to limit AI development, but to manage it responsibly. This is reported by Techcrunch.com reports .

In an article titled "A Foundation for Frontier AI and the Beginning of a New Era" published on his X account, Hassabis proposed introducing a system similar to the FINRA (Financial Industry Regulatory Authority) model in the US financial system. He believes such a body should test new models before they are released to the market and develop best practices for their deployment.

How would the new control mechanism work?

According to the proposed system, leading laboratories such as Google, OpenAI, and Anthropic would voluntarily submit their most powerful models to this standards body for review 30 days before public release. If the assessment protocol proves effective, this process could eventually become mandatory and a key requirement for deploying AI models in the US market.

Hassabis emphasizes that this body would work in collaboration with laboratories not only on pre-release checks but also on addressing critical vulnerabilities identified after the models are launched. This is a much more comprehensive approach than the ad hoc checks currently being conducted by the US government on Anthropic and OpenAI models.

According to ixbt.com, current government checks are being heavily criticized due to a lack of technical expertise and a lack of transparency in the decision-making process. The newly proposed organization would be supported by the government but funded directly by the AI industry and operate independently.

Balance between industry and government

The issue of regulating the AI sector remains a contentious topic between tech giants and the new US administration. Recently, White House AI advisor Sriram Krishnan stated that a separate AI oversight agency (like the FDA) would not be established within the executive branch. The self-regulatory organization model proposed by Hassabis could align with these political views.

Demis Hassabis envisions the following experts working in this body:

• Representatives of open-source software;
• Highly qualified technical experts from within the industry;
• Independent groups specializing in AI safety.

"The strength of this approach is that, while being technically sound, it supports innovation and encourages responsible behavior," writes the DeepMind head. According to him, the system is adaptable to rapid changes in the field and allows for the strengthening of control measures as the level of risk increases.

If implemented, this initiative could become the first real and effective global standard for AI safety. This would guarantee the reliability of AI systems not only for technology companies but also for ordinary users.