ทันเอไอ

Demis Hassabis, Google DeepMind CEO, Proposes Independent Frontier AI Regulatory Body Before Launch, Aiming to Set Safety Standards

The CEO and co-founder of Google DeepMind reveals a proactive plan, calling for the U.S. to establish an industry-funded AI regulatory body to inspect and certify the safety of the most advanced models before their public release.

Demis Hassabis, CEO and co-founder of Google DeepMind, has proposed establishing a new independent AI regulatory body in the United States. This body would have the authority to inspect the world's most advanced AI models and could coordinate a widespread development slowdown if dangerous risks are found. The proposal was detailed in his personal vision paper titled "A Framework for Frontier AI and the Dawning of a New Age."

Hassabis suggests this organization should be structured similarly to FINRA (Financial Industry Regulatory Authority), a private regulatory body funded by the industry itself to oversee Wall Street under government supervision. The proposed AI regulator would need a team of world-class technical experts and be accountable to the U.S. government. The process would require Frontier AI development companies to submit their models for safety inspection no more than 30 days before launch. They would need to pass dangerous risk tests before being deployed in the U.S. market.

Regarding the board structure, Hassabis envisions a majority of independent directors, including Turing Award recipients and recognized experts, alongside representatives from industry, government, and the open-source community. He also set a challenging timeline, stating it should take "a few months" and the organization should be operational "before the end of this year."

A key driver for this proposal stems from the Trump administration's sudden implementation of export controls on Anthropic's Mythos and Fable models last month. Hassabis called this a "significant wake-up call," demonstrating Washington's need for a more stable mechanism than ad hoc directives. He added that leaders from other major AI labs generally agree on the principle that the industry needs regulation, though details on who should hold regulatory power remain debated.