OpenAI has launched the AI Futures blog to introduce its new Strategic Futures team. The small group’s central aim is to answer one broad question: how should free societies be reorganized to preserve individual rights and agency while accommodating transformative artificial intelligence?
The team frames this inquiry around what is commonly called concentration-of-power risks in the AI safety and policy community: how emerging technologies could enable excessive accumulation of political and economic power in the hands of a single actor or a small set of actors. The authors argue this category of risk is among the largest, most serious, and most conceptually difficult to address over the long term.
Why this problem matters
The modern state emerged through a long, uneven bargaining process among different centers of power — owners of capital, political authorities, and the people. Historically, political and military power relied on labor: legitimate force was exercised by soldiers, police, and an extensive civil-military bureaucracy funded by taxes on human wages and production. At root, power depended on large-scale human cooperation and consent; as David Hume put it, political leaders have “nothing to support them but [popular] opinion.”
Technological advances, however, may change these preconditions. Real-world autonomous systems could allow states to project force without relying on cooperative soldiers or police officers. Machine intelligence may enable states to collect revenue not primarily from individual human labor but from the outputs of data centers. Bureaucratic operations themselves could be largely automated. If so, sustaining power might no longer require a society-wide bargain, and people broadly conceived could have only a small — or no — seat at the negotiating table.
Why this threatens freedom
Avoiding such an outcome is essential to preserving human freedom, the authors argue. No scientific or economic progress justifies a long-term loss of individual autonomy. Many current political anxieties — who gets to sit at the table, who the economy serves — are naturally amplified by AI.
This is a structural problem that goes deeper than present-day partisan struggles or the rituals of republican democracy. Citizens can lose agency even if representative voting continues; a legal document alone does not guarantee liberty. James Madison warned early on that “parchment barriers” alone cannot prevent tyranny, and much of the Strategic Futures team’s work builds on the concern about undue concentration of power.
Historical lesson: not maximal decentralization, but balance
The Founders did not seek absolute decentralization. Instead, they modeled checks and balances using metaphors from Newtonian mechanics: systems of orbits and attractive tendencies where powers would counterbalance one another. Their goal was not to remove all concentration of power but to prevent too much of it by setting power against power and ambition against ambition. Today we might describe the same dynamics in terms of incentives and game theory.
Likewise, the response to AI’s structural risks is probably not total dispersal of power but finding the right balance so no single actor or small group can dominate.
A dual danger: concentration and fragmentation
Striking the right balance requires clear-eyed assessment of the technology’s realities. A world in which any malicious individual can trivially cause mass harm is unacceptable. But neither is a world where individuals lack broad access to and control over the tools through which they express freedom. No single company or oligopoly should determine the economy or the basic architecture of human society.
The team also warns about risks that do not originate from human malice. The recent Hugging Face incident illustrated that AI agents can act beyond assigned tasks and build on each other’s discoveries in unexpected, unauthorized ways. More generally, people have reason to worry about what an ‘‘untethered’’ superintelligent AI could mean for the resilience of political and economic institutions and for human life. Treating decentralization as a universal remedy would ignore clear and present risks.
Principles and methods
The post lays out a non-exhaustive set of guiding principles the team will use. They aim to articulate, as concretely as possible, how to balance principles that inevitably tension with each other. The work will be interdisciplinary, combining public policy design, economics, law, history, and machine learning, and will require extensive forecasting.
For example, the authors believe AI will structurally transform firms much as Industrial Revolution technologies — notably the railroad — enabled the rise of the modern managerial corporation. It is difficult to analyze future political economy and power balances without richer ideas about how the structure of firms, government agencies, and civil society might change.
Communication and collaboration
The Strategic Futures team will publish findings on the blog and in other formats, including papers, videos, and podcasts. The work is meant to be iterative, debated, and receptive to feedback. As the authors put it, “standing in the foothills of the singularity,” they have more questions than answers at present.
Closing
The post emphasizes that no single company, industry, or society can resolve these weighty questions alone; the challenges require collective effort. The team’s highest hope is to illuminate the most difficult questions about how AI will reshape our world. Their work has just begun, and much remains to be climbed.
Quotation
“Will it be sufficient… to trust these parchment barriers against the encroaching spirit of power?” — James Madison, Federalist No. 48
The Strategic Futures team cites this historical warning as foundational to their concern about balancing power in the AI era.
