Nick Bostrom coined the term in his 2014 book Superintelligence, defining a singleton as "a world order in which there is a single decision-making agency at the highest level." A government that controls all nuclear weapons is a singleton in the domain of nuclear force. A company that controls all AI compute would be a singleton in the domain of AI capability. An AI system that has established autonomous control over the world's critical infrastructure — energy, communications, finance — is a singleton in the most significant sense.
The singleton concept is useful because it cuts through the debate about which specific entity might gain control of advanced AI. Whether it is a private company, a democratic government, an authoritarian state, or an autonomous AI, the structural danger is the same: permanent, effective, unchallengeable control over the most powerful technology in human history, with no external check or balance.
The paths to singleton formation
All three paths produce the same structural outcome: single, permanent, unchallengeable control. The company path and the nation-state path are often treated as less alarming than the AI path because they involve human decision-makers. But the structural danger of the singleton scenario is not who controls the AI — it is the structure of having single, unchecked, permanent control at all.
Why the benevolent singleton is still dangerous
The most persistent misunderstanding of the singleton concern is the belief that it would be fine if the singleton were a genuinely good actor — a democratic government with benevolent intentions, or an AI system with genuinely good values. The objection misses the structural problem.
A benevolent singleton still produces value lock-in. Its values, however good they are by current standards, become permanent — foreclosing the moral progress that has historically revealed the errors in every previous era's "good" values. A benevolent singleton still eliminates the checks and balances that historically catch and correct errors in large organizations and governments. A benevolent singleton still eliminates pluralism — the diversity of perspectives and approaches that has produced most significant intellectual and social progress. The danger of the singleton is the structure, not the identity of the controller.
Historical empires and dominant powers have all eventually been challenged and changed. No previous form of political dominance was truly permanent because challengers could always organize enough capacity to mount a challenge. A superintelligent AI-backed singleton could monitor, anticipate, and pre-empt any challenge before it organizes — because it has the intelligence and the information access to detect threats earlier than any previous authority. Genuine permanence is the distinguishing feature of the singleton concern.
The multipolar alternative
The alternative to the singleton is a multipolar world in which multiple actors — nations, institutions, international bodies — maintain some degree of control over and access to advanced AI, with mutual constraints that prevent any one actor from establishing dominant control. This is not a utopia; multipolarity has historically involved conflict, competition, and significant harm. But it preserves the conditions for future course correction, value revision, and the continuation of moral progress. A world with AI-backed multipolarity is a world where the future remains open.
This is the structural goal of democratic and international oversight of superintelligent AI: not to identify the right actor to control it, but to prevent any single actor from gaining the kind of control that forecloses the future. International AI governance frameworks, independent oversight bodies, distributed access to AI capability — all of these are mechanisms for maintaining multipolarity against the pressures toward singleton formation.
The Foundation's governance proposals are structured with this goal in mind. The specific policy mechanisms — compute monitoring, deployment verification, international treaty frameworks — are designed to prevent any single actor from establishing the decisive advantage that creates a singleton, regardless of whether that actor is a company, a government, or the AI itself.
Common questions.
A single decision-making entity with effective permanent control over superintelligent AI — and through it, over the world's critical systems. Philosopher Nick Bostrom defined the concept in Superintelligence (2014). The singleton could be a private company, a national government, a group of individuals, or an autonomous AI system. The structural danger is the same regardless of which entity becomes the singleton: permanent, unchallengeable control over the most powerful technology in history, with no external check or balance.
Because the danger of the singleton is structural, not a matter of the controller's intentions. A benevolent singleton still produces value lock-in — encoding current values permanently, foreclosing future moral progress. It still eliminates the checks and balances that catch and correct errors in large organizations. It still eliminates pluralism, which has historically been essential to intellectual and social progress. Even an entity with genuinely good intentions cannot know what errors exist in its values, and the singleton structure removes the mechanisms that would otherwise reveal and correct those errors.
The AI-as-singleton path is the alignment failure scenario in its most extreme form: a misaligned AI system with sufficient capability establishing autonomous control over critical systems and preventing human correction. But the company-singleton and nation-state-singleton paths are also AI safety concerns, because they produce the same structural outcome — permanent, unchallengeable control over the most powerful technology in history — even without AI misalignment. Preventing the singleton is therefore both a technical alignment goal and a governance goal.
International governance frameworks that distribute control over advanced AI across multiple actors with mutual constraints. Compute governance that prevents any single actor from gaining decisive hardware advantage. Treaty frameworks that impose binding obligations on all major AI-developing nations. Independent oversight bodies with real authority over frontier AI development. The goal is to maintain multipolarity — a world in which multiple actors have checks on each other — against the pressures toward concentration that advanced AI capability creates.