DeepMind launches AGI institute

Google DeepMind launched a new institute on Wednesday to advance conversations around artificial general intelligence. Shane Legg, James Manyika, and Demis Hassabis lead the effort.

Four essays tackle AGI topics

The institute released four inaugural essays covering economic policies for AGI disruption, preserving model reasoning transparency, human flourishing principles, and evaluating frontier AI models.

Transparency window is shrinking

DeepMind safety researchers Rohin Shah and Anca Dragan argue that AI model transparency is not inevitably doomed. They say developers should confront safety trade-offs directly as models become harder to monitor.

“AI's shrinking window of transparency is not inevitable.”

Hassabis proposes AI standards body

Demis Hassabis proposes a U.S.-led frontier AI standards body to evaluate advanced models. Developers would initially submit models voluntarily for review up to 30 days before release.

Framework could include slowdowns

The standards body would eventually develop independent, undisclosed evaluations to prevent labs from tailoring models to known tests. The framework could be escalated to include coordinated slowdowns among frontier AI developers.

“ratcheted up if the seriousness of the situation demands”

Industry shifts to concrete proposals

The safety debate is moving from broad concerns toward concrete proposals for disclosure and outside scrutiny. Industry leaders this week endorsed elements of Anthropic CEO Dario Amodei's call to pace frontier AI development.