Political and power concentration · Cybersecurity and infrastructure
Exclusive access to capabilities
A small group obtains decisive capabilities nobody else has, and with that stops needing anyone's permission.
- Severity
- Irreversible
- Horizon
- 3–10 years
- Evidence
- Projected
- Consensus
- Low
Of the three factors Davidson, Finnveden and Hadshar identify for an AI-enabled coup, this is the only one that rests on a present-day fact: frontier developmentFrontier 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 already confined to a few organisations led by a small number of people [420]AI-Enabled Coups: How a Small Group Could Use AI to Seize PowerView source ↗. The physical bottleneck confirms it from the other end —a sole producer of advanced packaging, three HBM memory suppliers sold out through 2026, four designers consuming 80-85% of supply— [375]AI Chip Supply Chain Bottlenecks and Capacity (AI Chip Components Explorer)View source ↗.
And tiers of access differentiated by capability, not just by price, already exist. Daybreak Red grants access to models trained for vulnerability research and exploitExploitThe specific program or technique that takes advantage of a security flaw to get into a system or take control of it.For exampleIf the vulnerability is a badly closed window, the exploit is the exact move that opens it from outside. development, with identity verification, monitoring, usage restrictions and physical security keys mandatory for all individual accounts from 1 September 2026 [822]Expanding Daybreak as the Cyber Defense Window NarrowsView source ↗archived copy only. That is access control, not capability control. The June 2026 episode shows the other hand on the same funnel: an export-control directive suspended access to two frontier models for every customer in the world, and the government first authorised restoration for a set of US organisations [85]Statement on the US government directive to suspend access to Fable 5 and Mythos 5View source ↗.
What this does not demonstrate. Restricted access is not decisive access, and the jump between the two is the projected part. The scale figure that underpins the argument —millions or billions of copies running year-round by 2030— is explicitly conditional on systems being as efficient as the human brain in operations per second, which is an assumption and not a measurement [420]AI-Enabled Coups: How a Small Group Could Use AI to Seize PowerView source ↗. The person who qualifies it most is its own author: in August 2025 Davidson says the dominant factor looks to him like singular loyalty in deployed systems, and that for exclusive access on its own “it’s a lot less clear” [106]46 - Tom Davidson on AI-enabled CoupsView source ↗.
Chain of materialisation
PreconditionObserved
Frontier development is already in few hands
It is the only one of the three AI-enabled coup factors resting on a present fact rather than a projection. The physical bottleneck confirms it from the other end: one exclusive advanced packaging producer, three HBM memory suppliers sold out, and four designers consuming 80-85% of supply.
TriggerObserved
Capability-tiered access levels already exist
Daybreak Red grants access to models trained for vulnerability research and exploit development, under identity verification, monitoring, usage restrictions and mandatory physical security keys since 1 September 2026. It is an access control, not a capability control: the same capability exists and only some may use it.
Precedents: OpenAI expands Daybreak and releases a variant trained for vulnerability research · An export control directive suspends access to Fable 5 and Mythos 5
Observed and demonstrated evidence ends here. What follows is projection.
CascadeProjected
The gap widens on its own if AI accelerates AI research
The projected argument is that if systems surpass human experts at AI research, the leading project accelerates and opens a gap. The scale figure they use -millions or billions of copies running year-round by 2030- is explicitly conditioned on an efficiency assumption nobody has measured.
ImpactSpeculative
Decisive capability with no counterweight
Davidson, who proposed the factor, is also the one who most qualifies it: in interview he says exclusive access on its own is far less clear than singular loyalty in deployed systems. It is the most speculative part of an already speculative literature.
Scenarios where it appears
See on the map →Report a mistake in this entry →
Sources
- [420] AI-Enabled Coups: How a Small Group Could Use AI to Seize Power · Forethought 2025
- [106] 46 - Tom Davidson on AI-enabled Coups · AXRP - the AI X-risk Research Podcast 2025
- [495] Extreme power concentration · 80,000 Hours 2025
- [375] AI Chip Supply Chain Bottlenecks and Capacity (AI Chip Components Explorer) · Epoch AI 2026
- [822] Expanding Daybreak as the Cyber Defense Window Narrows · OpenAI 2026 archived copy only
- [85] Statement on the US government directive to suspend access to Fable 5 and Mythos 5 · Anthropic 2026
- [83] Redeploying Claude Fable 5 · Anthropic 2026
- [367] What did US export controls mean for China's AI capabilities? · Epoch AI 2024
- [538] International AI Safety Report 2026 · International AI Safety Report (panel con representantes nominados por más de 30 países) 2026