If AI is becoming humanity’s most important technology, why are we building it like a smartphone app?
When I suggested that the world should stop building competing frontier AI models and instead create One AI – a single open-weight, open-source foundation model developed for humanity – I received the response I expected.
“It’s a nice idea, but it could never work.”
Could it?
History suggests otherwise.
Many of the technologies that underpin modern civilisation aren’t owned by a single company. They are governed through international collaboration, open standards and shared institutions.
The internet.
GPS.
The World Wide Web.
Linux.
The Human Genome Project.
Even CERN, where countries that disagree on almost everything still cooperate to advance science.
Perhaps the real question isn’t whether One AI could work.
It’s why we’re assuming the most powerful technology humanity has ever created shouldn’t be governed this way.
AI isn’t a product anymore
Today we still talk about AI companies as if they are software companies.
They’re not.
They’re building the next layer of global infrastructure.
Soon, AI will write software, discover medicines, operate factories, manage energy grids, educate children, conduct scientific research, secure computer networks and assist governments.
When something becomes this fundamental, ownership starts to matter.
Imagine if the internet protocol had been patented.
Imagine if email belonged to one corporation.
Imagine if every country had built its own incompatible version of the web.
We would never have achieved the connected world we have today.
Foundational infrastructure works best when everyone can build upon it.
AI should be no different.
One model. Millions of applications.
One of the biggest misunderstandings about One AI is that it would eliminate innovation.
Quite the opposite.
The proposal isn’t for one AI application.
It’s for one foundation model.
Think about Linux.
There isn’t just one Linux computer.
Thousands of companies build products on top of the same open operating system.
Red Hat.
Canonical.
Google’s Android.
Cloud providers.
Universities.
Research laboratories.
Everyone innovates.
Everyone competes.
But nobody wastes billions rebuilding the operating system from scratch.
One AI would work the same way.
The foundation model would be shared.
Everything above it would remain competitive.
Companies could still build:
- specialised healthcare assistants
- legal AI
- engineering co-pilots
- robotics platforms
- scientific discovery tools
- education systems
- creative applications
- enterprise products
Competition wouldn’t disappear.
It would simply move to where competition creates value.
How would it be governed?
The obvious fear is that One AI would become controlled by governments.
Or by one government.
Or by one company pretending to represent everyone.
That cannot happen.
Instead, governance would need multiple independent layers.
Technical governance
The model architecture, training code, datasets and evaluation results would all be public.
Every major update would be reproducible.
Researchers anywhere in the world could inspect the model.
Security researchers could audit it.
Universities could test it.
Companies could improve it.
Transparency becomes the default rather than an afterthought.
Scientific governance
Major changes would require independent review.
Not unlike scientific journals today.
Proposals would be published.
Evidence would be examined.
Performance would be evaluated openly.
The world’s researchers become participants rather than spectators.
International governance
Just as the Internet Engineering Task Force develops internet standards without belonging to any single nation, One AI would require an independent international foundation.
Countries would participate.
Universities would participate.
Industry would participate.
Civil society would participate.
No single organisation could quietly change humanity’s most powerful model.
Public governance
Most importantly, discussions about values would happen publicly.
Today we argue about AI alignment after companies have already trained their models.
One AI would reverse that.
Questions like privacy, bias, transparency and acceptable behaviour would become matters of public debate rather than corporate policy.
Open weights don’t make AI less safe
This is the objection people raise most often.
“If everyone has access to powerful AI, won’t that make the world more dangerous?”
It’s a fair question.
But it assumes secrecy is our strongest defence.
History suggests otherwise.
The internet’s security isn’t based on secret encryption algorithms.
Modern cryptography is public.
Linux is public.
Python is public.
The Apache web server is public.
OpenSSL is public.
Anyone can inspect the code.
Anyone can find vulnerabilities.
Anyone can contribute fixes.
The result isn’t perfect security.
It’s stronger security through collective scrutiny.
AI should adopt the same philosophy.
Open weights don’t guarantee safety.
But they make independent verification possible.
Closed models ask us to trust companies.
Open models allow us to verify them.
Those are very different things.
Funding One AI
Building frontier AI isn’t cheap.
Training the most capable models already costs billions of dollars.
But humanity is already spending those billions.
Just not together.
Instead, companies repeatedly spend enormous sums:
building nearly identical GPU clusters,
training overlapping datasets,
hiring the same researchers,
and pursuing the same benchmarks.
Much of this investment duplicates work that’s already been done elsewhere.
Imagine redirecting those resources.
Governments already fund large scientific infrastructure.
Universities already contribute world-class research.
Companies already invest hundreds of billions into frontier AI.
A shared foundation would reduce duplication while increasing transparency.
The investment doesn’t need to be larger.
It simply needs to be coordinated.
Defence becomes a shared responsibility
Perhaps the strongest argument for One AI isn’t economic.
It’s defensive.
If dangerous AI systems emerge – whether from criminals, rogue states or independent actors – humanity will need defensive AI capable of detecting, understanding and countering them.
That defensive system should not belong to one company.
Nor should every nation build its own competing version.
Cybersecurity already teaches us that defenders work best when information is shared.
Threat intelligence is exchanged.
Vulnerabilities are disclosed.
Security tools are developed collaboratively.
Advanced AI defence should follow the same model.
One AI becomes humanity’s immune system.
A common capability that protects everyone.
Transitioning without breaking innovation
No one expects every frontier lab to stop tomorrow.
The transition would happen gradually.
First, governments would prioritise funding open-weight research.
Then international research organisations would begin pooling compute resources.
Common evaluation standards would emerge.
Eventually, public investment would focus on a shared frontier model instead of funding fragmented national efforts.
Private companies would continue innovating – just as they do on top of Linux, the web and countless open-source technologies today.
The goal isn’t to abolish markets.
It’s to stop making intelligence itself a competitive arms race.
Why this matters now
Every year AI becomes more capable.
Every year it becomes more autonomous.
Every year it becomes more deeply embedded in society.
If we wait until superhuman AI exists before deciding how it should be governed, we’ll have waited too long.
The institutions that shape civilisation are built before they’re desperately needed.
Not afterwards.
We have a choice
Future generations may look back at this moment with disbelief.
They may wonder why dozens of companies raced to build increasingly powerful artificial intelligence in secret, competing for market share while asking society to trust them with civilisation’s most consequential technology.
Or they may see this as the moment humanity recognised that some things are simply too important to own.
The internet connected humanity.
Open science accelerated discovery.
Open-source software transformed computing.
Perhaps artificial intelligence is ready to become the next shared achievement.
One AI would not belong to Silicon Valley.
Or Beijing.
Or London.
Or Washington.
It would belong to all of us.
Not because that’s idealistic.
But because intelligence itself is becoming the infrastructure upon which everything else will depend.
The question isn’t whether AI will change the world. The question is whether the world’s most powerful intelligence will belong to a handful of companies, or to humanity itself.
2/3
First dropped: | Last modified: July 31, 2026