Preparing institutions for the transition to advanced AI.

AGI safety is not only about building safe AI models.

Many of the most consequential impacts play out across society, where only governments and shared institutions can act. Responsible scaling cannot be left to companies alone: it needs a layered, defence-in-depth response, like a Swiss cheese.

Extreme power concentration and catastrophic misuse must both be avoided.

Even as AI progress runs fast, no actor should gain a decisive strategic advantage over everyone else. The task is to manage the diffusion of advanced AI between two failure modes: dispersing capability widely enough to prevent domination by a single actor, while containing it enough to guard against catastrophic misuse and loss of control.

Our institutions for solving coordination problems were built for another era.

If we avoid catastrophic risks and a global AI dictator, we will still have many coordination problems in the future. However, the pacing problem is real: technology now moves faster than governance. We need more adaptive institutions that are able to sense, decide, and update quickly, including by leveraging advanced AI.