「Selling Out」 Google First Wants Ethics Board, 12 Years Later Hasebe Wants to Add Final Safeguard to AGI

According to Dynamic Beating Monitor, when DeepMind was sold to Google in 2014, one of Hassabis's conditions was to establish an independent AI ethics committee. Twelve years later, he is pondering the same thing, except this time he is not looking to oversee DeepMind but the entire cutting-edge AI industry.

The Wall Street Journal revealed that before stepping down as DeepMind's CEO, Hassabis had been working with senior U.S. government officials and other AI lab leaders to promote a new regulatory mechanism. U.S. Treasury Secretary Besent and President's Technology Advisor Michael Kratsios have both heard his proposal. Besent himself was involved in the Trump administration's AI strategy development. Hassabis likened the future institution to an AI version of the "International Atomic Energy Agency."

The subtle part is that the U.S. government has already taken the first step. In June, the White House introduced a voluntary review framework for cutting-edge models, and Hassabis's plan publicly released in July is remarkably similar even in operational details: both outline the cutting-edge models that require focused review, allow for up to 30 days of review before release, and involve government testing.

The disagreement lies only in the final step. The White House clearly stated that this mechanism should not turn into mandatory licensing or release approval; however, Hassabis wants to push further. Starting with voluntary participation, once the evaluation mechanism is established, it will become an entry barrier, and cutting-edge models that do not pass scrutiny will not be deployable in the U.S. Testing will also expand from network attacks to biological risks, agent bypassing security measures, and deceptive behavior, and if necessary, it can even coordinate several top labs to jointly slow down.