@mattbotvinick

I have been so happy to see concerns about "concentration of power" working their way into the AI safety agenda over the past couple of years. We ignored political risks for too long, and its great to see this now coming into focus. However, I think it's now time to go deeper, to focus on the preferred alternative to CoP. That means talking about how we assure a world in which we have separated powers in balanced equilibrium. There are two relevant models to work from: Acemoglu's and Robinson's balance between the power of the state and the power of society, and the Madison/Montesquieu model of separation of powers within the state. One challenge is to situate the power of the labs within the Acemoglu/Robinson model, and to do so not only descriptively but also normatively. I think we can make progress on that by starting from Philip Pettit's notion of non-domination as the goal of a liberal society. What is a good model for 'separated powers in balanced equilibrium' that (1) includes the labs, and (2) best promotes non-domination? --- ##### Comments > **fiona hackworth @fi0nahackworth** · [2026-09-06](https://x.com/fi0nahackworth/status/2096682958417457238) > > Truly see the need for this but would push back on the framing of labs’ role. Separation of powers is about constituted authority, and labs have none. Government is authority delegated by the people — no company can ever claim a seat at that table. > **Avi Siegal @AviSiegal** · [2026-09-06](https://x.com/AviSiegal/status/2096656353267781703) > > And let’s not forget the model of a geopolitical balance of power, à la the Congress of Vienna, Kissinger’s interpretation, etc. There the goal is not Pettit’s republican liberty but rather the comparatively more realistic goal of preventing war/catastrophe