Control has always followed infrastructure.
Those who controlled land shaped empires.
Those who controlled factories shaped nations.
Those who controlled finance shaped markets.
Today, control is migrating again—this time to artificial intelligence models.
The uncomfortable truth is that most public debates about AI regulation, safety, and ethics skip a more fundamental question:
Who actually controls the systems that increasingly think, predict, and decide on our behalf?
The answer is neither simple nor reassuring.
1. Control Is Not About Code — It’s About Leverage
Many assume that whoever writes the code controls AI.
That assumption is wrong.
Real control emerges from a combination of:
- Compute (who owns or can access large-scale hardware)
- Data (who can collect, aggregate, and retain it)
- Talent (who can hire and retain top researchers)
- Deployment channels (who integrates AI into daily life)
Control is not binary. It is layered.
You may use an AI model.
You may even customize it.
But that does not mean you control it.
2. Governments: Power Without Speed
Historically, governments controlled the most powerful technologies.
With AI, that dominance is eroding.
Most states:
- Do not own frontier models
- Do not manufacture advanced chips
- Do not operate hyperscale compute infrastructure
Even powerful governments increasingly depend on private AI systems for:
- Intelligence analysis
- Surveillance
- Logistics
- Cyber defense
Countries like the United States and China invest heavily in AI—but even they rely on private actors to execute.
Governments still wield:
- Legal authority
- Military budgets
- Regulatory power
But they lack something critical: iteration speed.
In AI, speed is power.
The Alignment Problem Is Not Technical — It’s Political
3. Corporations: Control Without Accountability
The most advanced AI models today are overwhelmingly developed and deployed by corporations.
They control:
- Training pipelines
- Model updates
- Access rules
- Pricing structures
- Usage constraints
This grants companies functional sovereignty over systems that influence:
- Communication
- Work
- Education
- Creativity
- Decision-making
Unlike governments, corporations are not bound by democratic legitimacy. Their incentives are:
- Profit
- Market dominance
- Competitive survival
This creates a paradox:
Some of the most powerful cognitive systems in human history are governed primarily by terms of service, not law.
And when corporate interests diverge from public interest, there is no clear override mechanism.
4. Regulation as a Proxy for Control
Many policymakers believe regulation equals control.
It doesn’t.
Regulation:
- Shapes behavior at the margins
- Sets constraints after deployment
- Rarely alters core power structures
The European Union has positioned itself as a global AI regulator. Yet it:
- Produces few foundational models
- Depends on external infrastructure
- Regulates systems it does not own
This creates regulatory asymmetry:
- Rule-makers without leverage
- Builders without oversight
Regulation can slow harm—but it does not transfer control.
👉 Related articles:
Why Regulation Will Always Be Late
5. The Illusion of Open AI
Open-source AI is often presented as the antidote to concentration.
In reality, openness has limits.
While open models:
- Increase transparency
- Encourage experimentation
They still require:
- Expensive compute to train
- Infrastructure to deploy
- Expertise to maintain
Open code without compute is not power—it is potential without execution.
As a result, even “open” AI ecosystems often consolidate around:
- Cloud providers
- Platform integrators
- Well-capitalized actors
Openness alone does not equal control.
The Myth of Neutral Algorithms
6. No One Controls AI — Or So It Seems
There is a third possibility: no one is fully in control.
AI systems are:
- Self-updating
- Interconnected
- Deployed across jurisdictions
- Integrated into countless workflows
Responsibility becomes fragmented:
- Developers blame users
- Users blame tools
- Governments blame markets
- Corporations blame demand
This diffusion of responsibility creates a governance vacuum.
When no single actor has full control, no one has full accountability.
7. Strategic Dependency: The New Vulnerability
As organizations integrate AI deeper into operations, a new risk emerges: strategic dependency.
Dependency looks like:
- Locked-in APIs
- Proprietary models
- Non-transferable workflows
- Opaque decision logic
Over time, users lose the ability to:
- Audit decisions
- Switch providers
- Operate without AI
At that point, control has quietly shifted upstream.
Not through force—but through convenience.
8. Control vs Influence: A Crucial Distinction
Few actors control AI outright.
Many actors influence it.
Influence comes from:
- Data generation
- Feedback loops
- Prompting and usage patterns
- Cultural norms embedded in interaction
But influence is not authority.
When models change:
- Users adapt
- Institutions adjust
- Societies normalize
The center of gravity remains elsewhere.
9. Why This Question Matters
Who controls AI models determines:
- Which values are embedded
- Which risks are prioritized
- Which voices are amplified
- Which futures are possible
Control over AI is not just technical—it is civilizational.
It shapes:
- Power distribution
- Economic outcomes
- Political agency
- Human autonomy
This is why the question cannot be deferred or abstracted.
10. The Emerging Reality
The most accurate answer today is unsettling:
AI is controlled by a shifting alliance of states, corporations, and infrastructure owners—without clear accountability or long-term stewardship.
This is not stable.
Systems that shape civilization require legitimacy.
Legitimacy requires transparency.
Transparency requires governance.
None are guaranteed.
Closing Thought
The danger is not that AI will be controlled by a single tyrant.
The danger is that it will be controlled by systems optimized for efficiency, profit, and speed—without anyone truly responsible for the consequences.
In such a world, power does not need to declare itself.
It executes—quietly, continuously, and at scale.