Skip to content
OGERIA — Observatory of Global Evidence on Risks in AISynthesis report · 2026 ed.
Updated 2 Oct 2026

Chapter 01 · Risk map

Political and power concentration · Military and autonomous weapons

Coup with singular loyalties

If military and administrative work is done by systems loyal to one person, seizing power stops requiring many people to go along.

Severity
Irreversible
Horizon
3–10 years
Evidence
Speculative
Consensus
Low

The argument of Tom Davidson, Lukas Finnveden and Rose Hadshar is not about technology but about the structure of power. Today, even dictators depend on others: military force requires personnel, government action depends on officials, production depends on a broad workforce. That dependence distributes power, and human reluctance to carry out illegal orders has historically been a brake. Their thesis fits in a single line: advanced AI removes that constraint, because it makes it technically feasible to replace human workers with systems singularly loyal to one person [420]AI-Enabled Coups: How a Small Group Could Use AI to Seize PowerDavidson, Tom; Finnveden, Lukas; Hadshar, Rose · 2025 · reportView source ↗Accessed on 9 September 2026.

The less obvious point is the second: singular loyalty not only enables the coup, it also cancels the response. Civil disobedience and the general strike work because human labour cannot be replaced; if even loyal supporters can be replaced, whoever seized power is left with no counterweight [420]AI-Enabled Coups: How a Small Group Could Use AI to Seize PowerDavidson, Tom; Finnveden, Lukas; Hadshar, Rose · 2025 · reportView source ↗Accessed on 9 September 2026. Finnveden, Riedel and Shulman reach the same point by another route: most regimes fall because their security forces defect, and that route closes when there are no soldiers who can defect [418]AGI and Lock-inFinnveden, Lukas; Riedel, Jess; Shulman, Carl · 2022 · reportView source ↗Accessed on 9 September 2026.

What this does not demonstrate. None of this is measured, and its authors say so: they analyse systems that surpass the best human experts in critical domains, not today’s AI, and present the exercise as conditional implications. In an August 2025 interview Davidson puts a number on his own thesis and disqualifies it in the same breath: “If I had to pluck a number, I’d say 10%, but it’s very made up” [106]46 - Tom Davidson on AI-enabled CoupsFilan, Daniel; Davidson, Tom · 2025 · interviewView source ↗Accessed on 9 September 2026. The strongest structural counterargument is an absence: in the extended summary for policymakers of the International AI Safety Report 2026 —more than a hundred experts, panels nominated by 29 countries— the word “concentration” does not appear even once [538]International AI Safety Report 2026Bengio, Yoshua · 2026 · reportView source ↗Accessed on 9 September 2026. This literature is specific and in the minority; that does not falsify it, but it places its status.

Chain of materialisation

  1. PreconditionObserved

    Power is distributed because people are needed

    To date, even dictators depend on others: military force requires personnel, government action depends on civil servants and economic output on a broad workforce. Between 1950 and 2010 there were over 200 successful coups, and a few battalions sufficed when they managed to keep other forces from intervening.

  2. TriggerObserved

    Loyalty becomes a parameter someone configures

    Today who obeys what is defined in usage policy documents written by a private company. The 2026 dispute shows the mechanism working and its fragility: Anthropic refused two use cases -mass domestic surveillance and fully autonomous weapons- on the basis of its own policy, and the state response was not to compel it but to exclude it from the federal supply chain.

    Precedents: The Pentagon-Anthropic dispute over autonomous weapons and surveillance

    Observed and demonstrated evidence ends here. What follows is projection.

  3. CascadeProjected

    The coalition requirement disappears

    Davidson, Finnveden and Hadshar argue that autonomous systems fully replacing soldiers could obey orders from a single person or small group, and that armed forces would be cautious but competitive pressures could lead to rushed adoption without adequate safeguards. They analyse systems surpassing top human experts in critical domains, not today's AI.

  4. ImpactSpeculative

    And the response to the coup disappears too

    The least obvious point of the argument is that civil disobedience and general strikes depend on human labour being irreplaceable. If even loyal supporters can be replaced, whoever took power faces no counterweight. It is a conceptual argument with no measurement possible today.

Scenarios where it appears

See on the map →Report a mistake in this entry →

Sources

  1. [420] AI-Enabled Coups: How a Small Group Could Use AI to Seize Power · Forethought 2025
  2. [106] 46 - Tom Davidson on AI-enabled Coups · AXRP - the AI X-risk Research Podcast 2025
  3. [418] AGI and Lock-in · Forethought 2022
  4. [495] Extreme power concentration · 80,000 Hours 2025
  5. [290] Pentagon-Anthropic Dispute over Autonomous Weapon Systems: Potential Issues for Congress · Congressional Research Service 2026 archived copy only
  6. [77] Where things stand with the Department of War · Anthropic 2026
  7. [73] Usage Policy · Anthropic 2025
  8. [299] Ukraine's Future Vision and Current Capabilities for Waging AI-Enabled Autonomous Warfare · CSIS Wadhwani AI Center 2025
  9. [538] International AI Safety Report 2026 · International AI Safety Report (panel con representantes nominados por más de 30 países) 2026

Ask OGERIA

It answers only with what the observatory publishes and can be wrong: check the entries it cites. Your questions are sent to an AI model, so don't write personal data. More in the privacy policy.

Up to 500 characters.

Support OGERIA on Ko-fi

The payment is processed by Ko-fi, not by this site. Open on ko-fi.com