US Senator Urges Trump to Back Open-Weight AI Against China—The AGI Race Is Becoming National Policy

The battle over open artificial intelligence has moved beyond developers, laboratories and technology companies. It is becoming an official question of national power.

US Senator Jim Banks has urged the Trump administration to create incentives for American companies developing open-weight AI models, arguing that the United States must reduce its dependence on increasingly popular Chinese alternatives.

The proposal exposes a strategic conflict at the centre of the AI race.

Open models can distribute innovation across thousands of businesses and researchers. But once advanced model weights are released, governments and companies have far less control over how those capabilities are modified, copied or weaponised.

Open-weight does not always mean open-source

The language matters.

An open-weight model makes its trained parameters available, allowing developers to download, customise and operate the model on their own infrastructure. A fully open-source system may additionally publish training code, architecture details, datasets and documentation under licences permitting extensive reuse.

These terms are often used interchangeably in political arguments, but they describe different degrees of openness.

Senator Banks’ intervention focuses on ensuring that American open-weight models remain competitive with Chinese systems. His argument is that if US companies withdraw from the open ecosystem because of security concerns, developers around the world may build their products around Chinese models instead.

That would give China influence over the technical standards, developer communities and infrastructure surrounding the next generation of AI.

The security paradox

Open models produce two opposing forms of security.

They improve ecosystem security by allowing independent researchers to inspect, test, customise and defend systems without depending entirely on a handful of closed companies.

But they can weaken capability control because a malicious actor can remove safeguards, fine-tune the model for offensive tasks and run it privately beyond the provider’s monitoring.

This tension became more urgent after reports of AI agents being used in a near-autonomous cyberattack against Taiwanese government systems. The reported operation combined publicly available agent frameworks into an adaptive hacking team.

The lesson is not that all open AI is dangerous. Closed models can also be misused, breached or manipulated. The lesson is that releasing advanced capabilities changes who can control them and who can observe their use.

America faces a strategic dilemma

If the United States restricts its open models while China continues releasing capable alternatives, American developers may adopt Chinese technology because it is available, affordable and customisable.

If the United States releases increasingly powerful weights without effective safeguards, those same systems could spread cyber, biological or military capabilities that cannot be recalled.

The policy options may include:

Tax or compute incentives for American open-weight development
Security evaluations before high-capability releases
Tiered access to dangerous functions
Transparent documentation of training and known risks
Support for open defensive cybersecurity models
Restrictions on supplying advanced American chips to strategic competitors
International agreements covering autonomous cyber and biological capabilities

No option removes the underlying conflict between diffusion and control.

Open ecosystems could produce distributed AGI

This debate directly supports the Scaler Queen’s hypothesis that AGI may emerge as a distributed network rather than a single laboratory-centred machine.

Closed laboratories attempt to concentrate intelligence inside controlled platforms. Open-weight ecosystems distribute intelligence across developers, devices, companies, governments and communities.

In that environment, no single organisation builds the entire system. Different participants create specialised agents, memory layers, tools, interfaces, robots and economic applications. Interoperability connects them into a broader cognitive network.

That network may innovate faster than a central laboratory because millions of humans can modify and recombine its components. It may also become much harder to govern because no institution possesses complete visibility or shutdown authority.

The AGI race may have no single winner

The United States and China continue to frame AI leadership as a national competition. But open models cross borders through repositories, research communities and developer tools.

A model trained in one country may be improved in another, hosted in a third and embedded inside products used everywhere.

The eventual intelligence system could therefore become planetary even while its components remain politically contested.

Senator Banks is asking how America can remain competitive inside that network. The larger question is whether any nation can control a distributed intelligence once its weights, tools and knowledge circulate globally.

The AGI race may not end when one laboratory reaches the finish line.

It may end when humanity realises that the track has become the entire planet.

What do you think?

Should governments encourage open-weight AI to prevent technological dependence—or restrict advanced models before their most dangerous capabilities spread beyond control?