Chaired by Demis Hassabis and directed by Shane Legg and James Manyika, the DeepMind Institute publishes four inaugural essays on model transparency and evaluation.
On September 17, 2026, Google DeepMind launched the DeepMind Institute, a structure designed to expose the differences of opinion between Google, DeepMind and the research community about artificial general intelligence. Demis Hassabis is the chairman, Shane Legg, co-founder of the lab, and James Manyika are directors, with Legg also serving as editor-in-chief.
Four inaugural essays open the work. They focus on the transparency of the reasoning of the models, on the proposal for a standardisation body for border systems led by the United States — initially on the basis of a voluntary review, possibly mandatory after approval — and on the frameworks for evaluating advanced systems. Economic policies to be countered to the expected upheavals are also on the agenda.
The initiative accompanies a shift in the debate on safety: from the general declaration of intent to concrete proposals for disclosure, external review and, for some, coordinated slowdown between laboratories.
Read the full article on the original source
Go to the source →