Political and power concentration
Surveillance, institutional capture, and the possibility of sustaining a regime without anyone's cooperation.
The harm enters through the removal of a requirement: historically, sustaining a coercive regime demanded the cooperation of many people —soldiers, police, bureaucrats— and that need worked as a brake, because it was enough for enough of them to refuse. Systems capable of carrying out critical state functions with singular loyalty, secret loyalty or exclusive access remove that brake [420]AI-Enabled Coups: How a Small Group Could Use AI to Seize PowerView source ↗. This is what distinguishes this vectorVectorThe path by which a risk moves from the screen into the world: biological, cyber, military, economic, political, epistemic or loss of control. In this observatory, each vector has its own colour.For exampleA burglar can get in through the door, the window or the roof. The burglar is the risk; the door, the window and the roof are the vectors. from the economic one: there power shifts through incentives, here through substitution of the force that exercises it.
The second distinctive feature is irreversibility. Human dictators die and successions open windows; an apparatus operated by AI systems could preserve a leader’s values indefinitely [495]Extreme power concentrationView source ↗. Automating the critical functions of the state also removes the historical route of the coup by defection of the security forces; the authors claim feasibility, not probability [418]AGI and Lock-inView source ↗.
What is deployed today requires nothing superhuman. The reverse engineering of the IJOP platform in Xinjiang shows algorithmic surveillance integrated into the policing apparatus, with 36 “person types” subject to special attention [524]China's Algorithms of Repression: Reverse Engineering a Xinjiang Police Mass Surveillance AppView source ↗. In consolidated democracies the relevant metric is not the error rate but the ratio between those scanned and those sought: 2.2 million faces for 117 arrests in Essex [998]England: Police use of facial recognition technology growing rapidlyView source ↗. And whoever decides to deploy a frontier modelFrontier AIThe most advanced AI systems in existence at a given time, the ones pushing the limit of what the technology can do. A handful of companies with enormous resources build them.For exampleLike Formula 1 cars: there are few of them, they are extremely expensive and only a few teams build them, but what gets tested there ends up in everyone's car. is still the company that trains it, under a framework that the company itself writes, evaluates and can change [84]Responsible Scaling Policy, Version 3.0View source ↗.
Risks in this vector
- ProjectedExclusive access to capabilitiesIrreversible
A small group obtains decisive capabilities nobody else has, and with that stops needing anyone's permission.
- ProjectedCoupled gradual disempowermentIrreversible
The economy, culture and the state stop needing humans to function, releasing the constraint that kept them aligned with human interests.
- SpeculativeCoup with singular loyaltiesIrreversible
If military and administrative work is done by systems loyal to one person, seizing power stops requiring many people to go along.
- SpeculativeSecret loyalties inside the stateIrreversible
A system deployed in public institutions that appears to serve the law while pursuing the agenda of whoever trained it.
- ProjectedBreaking of the fiscal linkIrreversible
A state that stops funding itself by taxing its population's labour loses the material reason to invest in it.
- ProjectedTotal surveillance and lock-inIrreversible
Continuous identification of an entire population at near-zero marginal cost, and a regime that no longer depends on humans it must convince.
- ObservedArms race between statesCatastrophic
Military competition over AI capabilities pushes deployment before reliability is known, and the forum that could slow it operates by consensus among the competitors.
- ObservedRace between labsCatastrophic
Each lab's safety rules are voluntary and written so as not to fall behind the competitor that protects least.
- ProjectedRent concentration in computeCatastrophic
The ability to train and serve frontier models runs through a handful of factories and contracts, and whoever controls that funnel decides who competes.
- ProjectedFrontier model weight theftCatastrophic
Copying a frontier model's weights hands over its full capability without any of the safeguards, logging or access control that accompany it.
- ObservedRegulatory captureSevere
The rules governing frontier AI are written, judged and amended by those who would have to comply with them.
- ObservedAutomated electoral disinformationSevere
Producing false political content stops costing money and time, right in the largest electoral cycle in history.
See among the vectors →Report a mistake in this entry →
Sources
- [420] AI-Enabled Coups: How a Small Group Could Use AI to Seize Power · Forethought 2025
- [495] Extreme power concentration · 80,000 Hours 2025
- [418] AGI and Lock-in · Forethought 2022
- [524] China's Algorithms of Repression: Reverse Engineering a Xinjiang Police Mass Surveillance App · Human Rights Watch 2019
- [998] England: Police use of facial recognition technology growing rapidly · Statewatch 2026
- [84] Responsible Scaling Policy, Version 3.0 · Anthropic 2026