gradual disempowerment

advancedoutlook

A vision in which humanity loses its say over its own fate not through a sudden machine uprising, but through many small, seemingly reasonable decisions.

In plain terms

In movies, artificial intelligence takes over power overnight. The idea of gradual loss of power is quieter: there is no coup and no evil machine, just year after year more decisions, work, and tasks are transferred to AI systems because they are cheaper, faster, and better.

It is like when the founder of a family business gradually hands everything over to managers. Each step is understandable. Yet ten years later, they notice their name is still on the company sign, but no one asks for their opinion, and they no longer see what is happening.

The essence of the theory is that the economy, the state, and culture today pay attention to people because they need them: our work, our taxes, our votes, our attention. If machines can also satisfy this need, institutions will have fewer reasons to serve people's interests.

The concept is useful for shifting attention from spectacular disasters to slow, systemic changes. These are harder to notice and, after a while, hard to reverse.

How it works

Companies and public offices are entrusting more and more tasks to AI systems and autonomous agents, because those who don't fall behind in the competition.
The economic weight of human labor is declining: an increasing share of income and capital is tied to machines rather than workers, weakening people's bargaining power.
The state becomes less dependent on its citizens: if most revenue and operations come from automated systems, there is less pressure to respond to public needs.
Culture and orientation are also transforming: more and more content, advice, and companionship are generated by machines, shaping what we think and want.
The areas reinforce each other, making them hard to correct: economic power becomes political influence, which further loosens regulations. In the end, humans formally still decide, but in reality only approve.

An everyday example

At your workplace, you used to evaluate incoming offers first, then approved AI suggestions, and now the system decides while you only see the weekly summary. No one took your authority away in a single step, yet you could no longer say why that particular supplier won.

Who is affected

everyoneThe question is whether you will retain real influence over how the economy, the state, and public discourse function in the long term.

employeeIf your work becomes less necessary, not only your salary but also your bargaining power will decline. → Work & HR

entrepreneur, executiveYou decide how much decision-making to delegate to machines and whether your company retains the knowledge needed to verify them.

lawyer, citizenDemocratic control depends on whether institutions rely on citizens and whether it remains transparent who or what makes the decisions. → Law

A common misconception

MythThe danger is that a self-aware, malicious AI might suddenly take control.

In realityFor this scenario, no evil machine or sudden twist is needed: it is enough to gradually trust more and more to well-functioning systems. Loss of power here is a byproduct of competition and convenience, not someone's plan.

Limits and risks

This is a theoretical scenario, not a forecast: no one can say what the probability is or how quickly it could unfold.
It is hard to measure: there is no accepted indicator for how much human influence has decreased, so the debate easily relies on guesses.
Researchers are divided: many believe that societies have previously adapted to major technological shifts, and that regulation and ownership structures can preserve human control.
The idea does not mean that automation is inherently bad. It warns that the sum of useful steps can still lead in a bad direction.

Similar, but not the same

sudden AI takeover of power – There, a system rapidly and intentionally seizes control; here, control slips from human hands slowly and unintentionally.
technological unemployment – This is only about job losses, whereas gradual loss of power also encompasses the erosion of political and cultural influence.
Automation bias – A single person's tendency to over-trust a machine is, in turn, a process that plays out at the level of entire societies and institutions.

Related concepts

Broader

existential risk, AI safety, AI risks

Narrower

loss of human oversight, outsourcing decisions to machines

On the edge

goal alignment, human oversight, Automation bias, superintelligence, technological unemployment

A brief history

The term was popularized by a study published in January 2025, authored by Jan Kulveit, Raymond Douglas, Nora Ammann, Deger Turan, David Krueger, and David Duvenaud. The authors emphasized that slow, systemic risks deserve special attention alongside sudden catastrophes. The idea is not without precedent: AI safety researcher Paul Christiano described a similar scenario of gradual decline in 2019.

Where things stand

As of late 2026, this is a contested research scenario, not an observed fact. It is regularly cited in AI safety and policy debates, but its measurement methods and countermeasures are still undeveloped.

Status: 2026 November

Also known as

gradual disempowerment · gradual disempowerment · gradual disempowerment · gradual loss of territory · the gradual marginalization of humanity · slow loss of power

Sources

This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.