Hassabis Proposes Mandatory AGI Safety Standards Modeled on IAEA Framework

DeepMind CEO Hassabis proposes a mandatory safety evaluation framework for advanced AI, aiming to transform voluntary US guidelines into binding entry requirements for model deployment.

Woofun AI reports that Demis Hassabis is proposing a regulatory framework for advanced AI, modeled after the International Atomic Energy Agency, before stepping down as DeepMind CEO. He has discussed these proposals with U.S. Treasury Secretary LScott Bessent and President's Technology Advisor Michael Kratsios. The plan mirrors the White House's voluntary review framework but seeks to convert it into a mandatory licensing requirement, preventing deployment of models that fail safety standards. Testing protocols would expand to cover biological risks and deceptive behaviors, potentially involving coordination among top laboratories to delay releases if necessary.