OpenAI's new policy team says the AI risk is a state that no longer needs its citizens

OpenAI has staffed a new policy unit, Strategic Futures, around a thesis that is older than the lab itself: the political theorist's worry that whoever commands coercive force sets the terms of social life, applied to a future in which force is software. The team, led by former Trump White House AI adviser Dean Ball, makes a deliberate bet that the most consequential AI risk is not a misaligned machine but a state empowered by autonomous systems to do what it wants without the people it governs.

The team's opening post, "Introducing AI Futures," frames the program under a "concentration of power risks" lens, a subcategory of AI safety and policy that asks what happens to individual rights when governments can project force and collect revenue without asking. In the team's own phrasing, "advancements in real-world autonomous systems may soon mean that states will be able to project force—domestically and internationally—without relying on cooperative soldiers or police officers." The post leans on James Madison's Federalist No. 48, on the tendency of power to accumulate unless actively dispersed, and on David Hume on popular opinion as the substrate of political authority.

That is a different pitch from the framing OpenAI and much of the frontier-AI world have used in public since 2023. The dominant story has been extinction and misalignment, the worry that a capable model will pursue goals its creators did not intend and treat humanity as collateral. Strategic Futures argues, in effect, that even a perfectly aligned model changes the balance of power between citizen and state, and that designing the political answer should start before the technology lands. As Ball put it on the Cognitive Revolution podcast, he wants to be in the room where "new power centers" and "frontier AI policy" meet, with what he calls "main character energy."

The argument has already cost the team goodwill in the room it most needs. A top Pentagon official publicly targeted Ball in mid-July, per Axios, in a fight that has been read as friction over the line between civilian AI policy and defense applications. OpenAI is publicly placing a Madison-shaped bet about state power, and the state is responding.

Ball arrives with standing to make the bet. A Hoover Institution fellow before joining OpenAI, he helped shape the Trump administration's AI executive order work and has spent the past year writing a Substack widely read in the AI policy world. A Business Insider profile describes him as "OpenAI's Substacker-in-chief," a label the framing of the launch seems designed to honor. The launch post itself is a single author's view rather than an OpenAI corporate position, and it should be read that way; OpenAI's other policy work, including the Frontier Safety Blueprint, still travels under the more conventional technical-risk framing.

Whether the Madison frame travels beyond Ball's blog is the open question. The "concentration of power" lens has critics inside the AI safety community, who read it as a way to redirect attention from technical alignment toward a political project the field has not signed on to. Civil-liberties groups have historically focused on surveillance, procurement, and deployment, not on the philosophical scaffolding around state power; labor groups have focused on displacement, not on consent. If Strategic Futures wants to set the terms of the next policy cycle, it has to convert the philosophical bet into something each of those communities will sign.

The first posts will tell. A team blog is a content channel until the people reading it in government and civil society treat it as a brief. Strategic Futures is betting that the next AI fight is not a fight about AI at all, but about who is left in the room when the state no longer needs to ask.