Commentary. The politics of artificial intelligence is entering a more dangerous phase. Governments are no longer debating whether advanced AI matters; they are deciding who gets access to the computing power, chips and infrastructure that make it possible. That shift is understandable. It is also a test of whether policymakers can distinguish real security risks from the temptation to treat technological competition as a zero-sum war.

Recent developments show why the issue cannot be reduced to slogans. Reuters reported that scrutiny of Nvidia chips reportedly appearing in China has revived questions about whether Washington’s export controls are working as intended.[1] Separately, a Reuters analysis described artificial intelligence as part of a wider group of fast-moving risks—including cyber disruption, drones and vulnerable infrastructure—that is reshaping governments’ assumptions about security.[1]

Our view is straightforward: advanced AI requires firm guardrails, but broad technological containment is unlikely to deliver lasting security. It may instead encourage evasion, accelerate fragmentation and weaken the international cooperation needed to manage systems whose effects cross borders.

The case for controls

There is a serious argument for restricting access to the most capable chips and models. Advanced computing can support military planning, cyber operations, surveillance and the rapid development of autonomous systems. Governments also have legitimate reasons to prevent sensitive technologies from strengthening hostile states or enabling dangerous non-state actors.

Controls can buy time. They can make it harder to obtain the hardware needed for the most resource-intensive applications, give regulators leverage over companies and signal that commercial access is not more important than public safety. In an unstable world, doing nothing would not be neutrality; it would be a policy choice.

That case deserves respect, particularly when technological developments are arriving faster than traditional security institutions can assess them. The Atlantic Council’s recent analysis of the Trump-Xi-Putin summit likewise reflects a broader concern: strategic competition is increasingly being managed through encounters among great powers, while smaller countries face the consequences of decisions they did not shape.[2]

Why blanket restriction is the wrong answer

Yet export controls have limits. Hardware can be rerouted, stockpiled or replaced by domestic alternatives. Restrictions can also push research into less transparent environments, where safety testing is weaker and international communication is thinner. If every advanced chip becomes an instrument of geopolitical denial, companies and researchers will have incentives to build parallel ecosystems rather than common standards.

There is an economic cost, too. AI is already being integrated into health care, scientific research, manufacturing and public services. Small firms and universities may be unable to absorb compliance burdens that large corporations can manage. A policy intended to protect societies could therefore concentrate power in the biggest technology companies while limiting beneficial experimentation.

Critics of a lighter-touch approach will rightly say that innovation does not automatically produce safety. They are correct. Voluntary promises have repeatedly proved insufficient when commercial rewards are immediate and accountability is diffuse. But that does not make indiscriminate restriction effective. It means regulation should focus on measurable capabilities and uses: controls on the most dangerous systems, mandatory testing and reporting, clear liability rules, and independent oversight.

A more durable bargain

Policy should begin with transparency. Governments should explain which capabilities trigger restrictions and publish evidence that those restrictions are achieving their stated goals. Companies should be required to report serious incidents, evaluate high-risk models and protect critical infrastructure from foreseeable misuse. Allies should coordinate rules rather than impose overlapping regimes that create loopholes and confusion.

Most importantly, democratic governments should preserve channels for scientific and diplomatic contact. Competition is unavoidable, but secrecy without communication increases the chance of miscalculation. Shared standards for testing, incident reporting and crisis communication would not eliminate rivalry. They would make rivalry less reckless.

The alternative is a cycle of escalation in which each new model is treated as proof that the previous controls failed, prompting ever broader restrictions. That approach may satisfy political demands for action while leaving the underlying risks unresolved.

AI policy should be neither naïve nor theatrical. Security matters; so does openness. The goal is not to win a permanent race to deny others access to technology. It is to ensure that the most powerful systems remain governable, accountable and compatible with public safety. That will require restraint from companies, precision from governments and cooperation from countries that increasingly distrust one another.

Sources: Reuters commentary on cyber, drone and AI risks; Atlantic Council analysis of great-power relations; European Council prospects for 2026.