Third Risk

The central argument is that AI could pose a “third risk” even if advanced systems are successfully aligned with their owners: a small number of private actors could gain control of machines capable of doing nearly all economically valuable work. Recursive improvement in software and robotics could make that control increasingly powerful, while ordinary people lose the labor and bargaining power that historically forced rulers to meet their needs. The result could be “AI feudalism”: owners need not provide much beyond cheap entertainment or stipends, yet their demand for land, energy, water and minerals could displace everyone else. This outcome is not inevitable, but the piece argues that human control of AI is not the same as democratic control over who wields it.

It also describes how lab leaders might shift from pursuing profits to securing the assets, talent and infrastructure needed to entrench their power, while presenting concern about AI safety as a shared priority. Competitive pressure and weak transparency could conceal such a shift without any explicit conspiracy. An analysis of 393 public reactions to three major AI stories found that power concentration received little sustained attention: public discussion focused more on conventional governance, safety or skepticism, while 51% of 88 expert works examined the issue. The piece argues that both AI believers and skeptics can miss the political stakes - whether they focus on alignment failures or on the possibility of an investment bubble. Its key distinction is between asking whether AI can be controlled and asking who controls it.