When Capability Spreads, Does Control Follow?
One idea has shaped much of the conversation around AI.
As powerful models become more accessible, control should become more distributed.
It is an intuitive assumption.
More organizations can build.
More developers can innovate.
More capability becomes available across the ecosystem.
But capability and control are not necessarily the same thing.
That distinction is becoming harder to ignore.
Across today's AI landscape, models continue to improve and access continues to broaden.
Yet many of the decisions that ultimately shape outcomes appear to be concentrating somewhere else.
A model may remain unchanged.
Its deployment conditions may not.
An application may use exactly the same model.
Its default behaviour may change overnight.
Access policies can evolve.
Interfaces can shape user behaviour.
Distribution channels can determine which systems are adopted at scale.
None of these require changing the underlying model.
Yet each can still change outcomes.
That raises a different question.
Perhaps the challenge is no longer understanding how capability spreads.
It is understanding where outcome-changing authority ultimately resides,
once AI systems become increasingly layered.
Why Existing Mental Models Become Insufficient
When we discuss AI control, we often assume authority can be located.
Some believe it resides with model developers.
Others point to compute infrastructure.
Some focus on regulation.
Others argue that applications or distribution channels ultimately determine outcomes.
Each perspective captures part of the system.
None fully explains how the system behaves.
The reason is subtle.
Each assumes authority is relatively static.
Modern AI systems are not.
They have become increasingly layered.
Models are only one component.
Infrastructure, orchestration, interfaces, enterprise workflows and distribution channels all influence outcomes.
As these layers evolve together, the authority to change outcomes does not necessarily remain where capability originated.
It can migrate.
Once that becomes possible, identifying the most capable layer no longer tells us where meaningful authority resides.
The analytical challenge changes.
It is no longer enough to ask:
Who owns the capability?
We must also ask:
Which layer can still change outcomes?
That distinction creates an important explanatory gap.
Existing governance models explain responsibility.
Existing architecture diagrams explain system design.
Neither systematically explains how outcome-changing authority behaves as layered AI systems evolve.
Authority Migration: A Structural Property of Layered AI Systems
If existing mental models cannot fully explain where authority resides,
then the missing question is not who controls AI – but how authority behaves as AI systems evolve.
One recurring structural pattern helps explain that behaviour.
This research refers to that pattern as Authority Migration.
 |
| Authority Migration across layered AI systems |
Authority Migration is the movement of outcome-changing authority across system layers, as AI ecosystems evolve and new dependencies emerge.
The important point is not that authority disappears.
It does not.
Nor does it simply transfer from one layer to another.
Instead, different layers become more influential as technical, commercial and institutional conditions change.
Why does this happen?
Because authority tends to migrate towards the layers that become progressively harder to replace, bypass or operate without.
As AI ecosystems mature, new dependencies emerge.
Enterprise workflows become embedded.
Interfaces shape behaviour through defaults.
Orchestration frameworks coordinate increasingly complex tasks.
Distribution ecosystems influence adoption at scale.
Capability may become more widely available.
Dependency often does not.
As dependencies accumulate, authority increasingly settles where meaningful intervention remains possible.
That has an important implication.
Ownership alone no longer explains control.
Model performance alone no longer explains influence.
Even openness, by itself, no longer explains where meaningful authority resides.
Authority is better understood as a dynamic property of the system rather than a fixed property of any single layer.
Recognising that shift changes the question entirely.
If authority can migrate, how do we systematically identify where it resides?
Recognising Authority Migration explains the phenomenon.
It does not yet provide a systematic way to analyse it.
That is the gap Control Topology is designed to address.
Introducing Control Topology
Neither ownership models nor architecture diagrams, on their own,
can systematically identify where outcome-changing authority resides.
What is missing is a way to systematically analyse authority itself.
This research introduces Control Topology to address that need.
Control Topology is an analytical framework for identifying where outcome-changing authority resides within layered AI systems, how that authority migrates over time, and which layers ultimately determine outcomes.
That distinction matters.
Authority Migration describes a structural phenomenon.
Control Topology provides the analytical framework for understanding it.
The framework does not prescribe how AI systems should be designed.
It does not recommend where authority ought to reside.
Instead, it asks a different question.
Under current conditions, which layer can still meaningfully change outcomes?
That shifts the discussion from description to diagnosis.
AI architecture explains how a system is constructed.
Control Topology explains how authority behaves within that system.
Those are fundamentally different questions.
One describes structure.
The other evaluates influence.
That distinction becomes increasingly important as AI ecosystems continue to evolve.
Mapping Before Predicting
Much of today's discussion focuses on predicting where AI power will concentrate.
Will foundation models dominate?
Will infrastructure providers?
Will enterprise platforms?
Will regulators?
These are important questions.
But they come later.
Before we can predict where authority might move, we first need a reliable way to identify where it resides today.
Without that, prediction risks confusing capability with control.
Control Topology therefore begins with diagnosis.
It asks:
• Where meaningful intervention remains possible?
• Which dependencies have become hardest to replace?
• Which layer ultimately determines outcomes?
Only then does it become meaningful to examine how authority may evolve in the future.
Prediction without diagnosis risks confusing capability with control.
Understanding where outcome-changing authority resides is the first step toward understanding how increasingly complex AI systems actually behave.
The First Law of Control Topology
Authority Migration explains that outcome-changing authority can move across system layers.
The next question is whether that movement follows a recognisable pattern.
If authority migrated randomly, mapping it would have little practical value.
Control Topology therefore begins with a more fundamental observation.
Authority does not migrate without direction.
It tends to follow changing patterns of dependency.
This research refers to that recurring pattern as the Topological Law.
The Topological Law states:
Capability tends to diffuse.
Dependency tends to accumulate.
Authority tends to migrate.
Control ultimately settles where meaningful intervention remains possible.
Each part of the law describes a different property of layered AI systems.
They should not be treated as interchangeable.
Capability often spreads first.
New foundation models become available.
New applications emerge.
More organisations gain access to advanced AI capabilities.
Innovation expands across the ecosystem.
This creates the impression that control is becoming equally distributed.
But another process unfolds at the same time.
Dependencies begin to accumulate.
Applications integrate with shared infrastructure.
Enterprises standardise workflows.
Users adapt to familiar interfaces.
Developers build around common orchestration frameworks.
Each individual dependency may appear insignificant.
Collectively, they reshape the system.
Over time, replacing those dependencies becomes increasingly difficult.
Switching costs rise.
Coordination becomes more complex.
Intervention becomes concentrated in fewer places.
That is when authority begins to migrate.
Not because capability has disappeared.
But because certain layers acquire a greater ability to influence outcomes than others.
Eventually, meaningful control settles where intervention remains possible.
That layer is not predetermined.
It may be infrastructure.
It may be orchestration.
It may be an interface.
It may be a distribution ecosystem.
The framework does not assume where authority should reside.
It asks where meaningful intervention actually remains possible under current conditions.
This distinction is important because many discussions assume openness naturally distributes control.
The Topological Law suggests something different.
Capability and authority often evolve in different directions.
One becomes broader.
The other becomes increasingly shaped by dependency.
That is why Control Topology begins by mapping dependencies rather than capabilities.
Capabilities explain what a system can do.
Dependencies explain which layer can still change outcomes.
That distinction is what gives the Topological Law diagnostic value.
Why Dependencies Matter More Than Capabilities
Traditional analysis often starts with technical capability.
Which model performs best?
Who trained it?
Who owns it?
These questions remain important.
But they do not always explain who can still alter outcomes after deployment.
Control Topology therefore asks a different set of questions.
• Which layer has become hardest to replace?
• Which dependency can no longer be bypassed?
• Where does meaningful intervention still exist?
The answers need not point to the same layer.
That is precisely why dependency deserves separate attention.
Capability can spread rapidly.
Dependencies usually compound slowly.
Yet it is often the slower process that determines where durable authority eventually resides.
This observation also explains why Control Topology is not another architecture model.
Architecture describes relationships between system components.
Control Topology evaluates how those relationships affect outcome-changing authority as the system evolves.
The objective is not to identify the most important layer.
It is to understand how changing dependencies alter where meaningful influence ultimately resides.
From Mapping Authority to Diagnosing Systems
The Topological Law provides an important insight.
It explains why authority tends to move across layered AI systems.
It does not, however, determine where authority resides within any individual system.
Different systems accumulate different dependencies.
Different architectures create different intervention points.
Different ecosystems concentrate authority in different ways.
Understanding that variation requires something more systematic.
The next step is not another principle.
It is a diagnostic method.
Control Topology therefore moves from recognising a general law to evaluating individual systems.
Only then can authority be mapped consistently across different AI ecosystems.
The Control Topology Diagnostic
The Topological Law explains why authority tends to migrate.
The next challenge is practical.
How do we identify where authority resides within a specific AI system?
Looking at a single layer is rarely enough.
Modern AI ecosystems consist of multiple technical, commercial and institutional layers.
Each contributes differently to system behaviour.
But not every layer possesses the same ability to change outcomes.
Control Topology therefore begins with a simple principle.
Every layer should be evaluated by the authority it can still exercise,
not by the capability it originally created.
That distinction changes the analysis.
A layer may contain the most advanced technology.
Yet another layer may retain the greater ability to influence deployment, access, behaviour or adoption.
Capability and authority can therefore become separated.
The purpose of the Diagnostic is to identify that separation.
Rather than asking where AI capability originated, it asks a different set of questions.
• Which layer can still alter outcomes?
• Which layer has become hardest to bypass?
• Which layer continues to shape behaviour after deployment?
Those questions form the basis of the Control Topology Diagnostic.
The Diagnostic does not begin with ownership.
It begins with intervention.
Every layer is first evaluated through one foundational question:
Can meaningful outcomes still be changed from here?
If the answer is yes, that layer retains authority.
If the answer is no, capability alone is no longer sufficient to explain control.
 |
| Control Topology Diagnostic for Layered AI Systems |
A Simple Illustration
Consider a foundation model that remains technically unchanged.
Its access conditions become more restrictive.
An application updates its default behaviour.
Enterprise workflows adapt to those defaults.
Users experience different outcomes.
The underlying capability has not changed.
But the authority to influence outcomes has.
Control Topology therefore does not ask which layer is most important.
It asks which layer retained the ability to change outcomes.
That distinction becomes increasingly important as AI systems mature.
Authority therefore cannot be inferred from capability alone.
The Diagnostic is therefore designed to be dynamic.
It captures where authority resides today.
Not where it originated.
Nor where it may reside in the future.
That makes Control Topology different from conventional architecture analysis.
Architecture explains how components connect.
The Diagnostic evaluates which of those components can still influence outcomes as conditions change.
In that sense, Control Topology is less concerned with how a system is built than with how authority behaves once the system is operating.
Looking Beyond Individual Layers
The Diagnostic also avoids another common assumption.
It does not assume that one layer permanently controls the entire system.
Authority may be distributed.
It may overlap.
It may migrate again as new dependencies emerge.
The objective is therefore not to identify a permanent centre of control.
It is to identify the layer or combination of layers that currently retains the greatest ability to influence outcomes.
That distinction matters.
Static diagrams describe systems.
The Diagnostic evaluates systems that continue to evolve.
For layered AI ecosystems, that difference is becoming increasingly important.
Why This Matters Before We Examine Real Systems
A framework should do more than describe the world.
It should improve how we interpret it.
Without a consistent diagnostic, different observers often reach different conclusions about where authority resides.
Some focus on models.
Others on infrastructure.
Others on regulation or enterprise adoption.
Each may be observing a genuine source of influence.
But each may also be observing only part of the system.
Control Topology does not begin by asking which perspective is correct.
It begins by asking whether they are analysing the same layer.
Only after authority has been mapped consistently does comparison become meaningful.
That provides a common analytical foundation for evaluating AI ecosystems that continue to evolve.
A diagnostic is only the starting point.
It identifies where authority appears to reside.
Evaluating that diagnosis consistently requires a common set of analytical dimensions.
What the Market Is Actually Signaling
 |
| Perceptions of AI Control |
The poll produced an interesting result.
There was no clear consensus.
Responses were spread across every option.
That matters more than any individual outcome.
It suggests readers instinctively searched for authority in different parts of the AI system.
Some looked at model developers.
Others looked at infrastructure.
Some focused on governments.
Others pointed to organisations deploying AI.
Each perspective is understandable.
Each identifies a genuine source of influence.
Yet the absence of consensus reveals something deeper.
We continue to look for authority as though it resides in one place.
The Control Topology Diagnostic begins from a different premise.
Authority is not necessarily a fixed property of any single layer.
It is a dynamic property of the system.
As dependencies evolve, authority can migrate.
That means different observers may all be correct about different parts of the system while still reaching incomplete conclusions about where outcome-changing authority ultimately resides.
The poll does not identify the "right" answer.
It demonstrates why a new analytical framework is needed.
Without a consistent way to map authority, equally informed observers can interpret the same AI ecosystem through entirely different lenses.
The objective of Control Topology is not to decide which layer matters most.
It is to provide a consistent method for determining which layer or combination of layers retains meaningful authority under current conditions.
That distinction becomes increasingly important as AI ecosystems continue to evolve.
Why It Matters
Investors
The question is no longer limited to which models are becoming more capable.
It increasingly becomes which system layers are accumulating durable authority as dependencies grow.
Control Topology shifts attention from capability leadership to the system layers where durable influence and long-term value may ultimately concentrate.
Policymakers
Governance is often designed around clearly identifiable centres of authority.
Layered AI systems make that assumption increasingly difficult to sustain.
Understanding where meaningful intervention remains possible may become just as important as identifying who is formally responsible.
Builders
Technical capability alone does not determine long-term architectural leverage.
Choices around orchestration, interfaces, workflows and distribution can gradually reshape where authority resides.
Understanding those dependencies becomes increasingly important as AI ecosystems mature.
Risks, Constraints & Open Tensions
Control Topology provides a way to diagnose where authority resides.
It does not eliminate uncertainty.
Layered AI systems continue to evolve.
So does the distribution of authority within them.
Several questions therefore remain open.
Authority Can Migrate Faster Than Governance
Governance frameworks often assume authority can be identified once and regulated accordingly.
Control Topology suggests that assumption may not always hold.
As dependencies evolve, outcome-changing authority may migrate to different system layers.
Governance therefore becomes a continuous process of diagnosis rather than a one-time exercise.
Authority May Become Distributed Rather Than Centralised
The framework should not be interpreted as assuming authority always concentrates into a single layer.
In some systems, meaningful authority may remain shared across multiple layers.
Different actors may each retain the ability to influence different outcomes.
The analytical task is therefore not to identify one permanent centre of control.
It is to understand how authority is distributed under current conditions.
Diagnosis Depends on Changing Conditions
A Control Topology diagnosis is not permanent.
New interfaces emerge.
Dependencies strengthen or weaken.
Enterprise adoption changes.
Regulatory intervention alters incentives.
As the system changes, authority may migrate again.
Mapping authority therefore cannot be treated as a one-time assessment.
Mapping Authority Does Not Predict the Future
Control Topology is designed to identify where authority resides today.
It does not claim to predict where authority will ultimately settle.
Prediction becomes more meaningful only after authority has first been mapped consistently.
Understanding the present remains the foundation for reasoning about future change.
Control Topology Evaluation Metrics
The Control Topology Diagnostic identifies where authority appears to reside.
The next question is equally important.
How should that diagnosis be evaluated consistently?
Different observers may examine the same AI system and still reach different conclusions.
The Diagnostic identifies where authority appears to reside. It does not, by itself, provide a common standard for evaluating that diagnosis.
Control Topology therefore completes the Diagnostic by introducing a common set of evaluation dimensions.
These metrics are not designed to measure technical performance.
They are designed to evaluate where outcome-changing authority ultimately resides.
1. Intervention Rights
Who can still halt, redirect or override outcomes?
The ability to intervene often reveals more about authority than the ability to create capability.
2. Dependency Concentration
Which layer has become hardest to replace or bypass?
Durable dependencies often indicate where authority is becoming embedded.
3. Outcome Influence
Which layer can alter behaviour at the greatest scale?
Small technical changes can produce disproportionately large system-wide effects.
4. Value Capture Alignment
Does the distribution of capability produce a similar distribution of economic value?
Or has value accumulated somewhere else?
Comparing capability with value helps reveal whether authority is migrating.
5. Time-to-Recentralization
After capability becomes more widely available, how quickly does authority begin to concentrate elsewhere?
This highlights the speed at which new dependencies emerge.
Bringing the Framework Together
The Topological Law explains why authority migrates.
The Diagnostic identifies where authority currently resides.
The Evaluation Metrics provide a consistent way to assess that diagnosis across different AI systems.
Together, they form the analytical framework for systematically mapping outcome-changing authority within layered AI systems.
Rather than asking who owns AI, Control Topology asks a different question.
Where does outcome-changing authority reside under current conditions,
and how might it migrate as the system evolves?
A New Way to See AI Systems
Every major technology wave changes the questions we ask.
The early internet changed how we thought about information.
Cloud computing changed how we thought about infrastructure.
AI is changing how we think about authority.
For many years, we have searched for control by asking who owns the models.
Who built them?
Who regulates them?
Who deploys them?
Those questions remain important.
But layered AI systems increasingly demand another question.
Where does outcome-changing authority actually reside?
Control Topology does not suggest that authority belongs to one particular layer.
Nor does it claim that authority will always migrate in the same direction.
Instead, it offers a different way of analysing AI systems.
It begins by recognising that authority is not a fixed property.
It is a dynamic property that can migrate as dependencies evolve.
That shift changes more than our understanding of AI.
It changes how we approach governance.
How we evaluate investment.
How we design systems.
How we identify meaningful intervention.
Most importantly, it changes what we look for.
The question is no longer who created the capability.
It is who can still change the outcome.
Perhaps the future of AI will not be determined solely by how widely capability spreads.
It may instead be determined by
how well we understand where outcome-changing authority resides.
Building on this Research
Applied Case Study #1: Mapping the Control Topology of the OpenAI Ecosystem
Further publications coming soon.
P.S. Original research by AI Block Assets Hub™
Author
Indrajit Chakraborti
Researcher & Founder – AI Block Assets Hub™
AI Block Assets Hub™ publishes original, decision-grade research at the intersection of AI, Blockchain, and Digital Assets.
Comments
Post a Comment