The way leaders talk about artificial intelligence is becoming part of the technology’s future. Verity Harding, a former head of global public policy at Google DeepMind, argues that the popular idea of an AI arms race may be steering governments and companies toward the wrong choices.
Harding’s concern is not only that the phrase sounds dramatic. Her argument is that language can narrow the policy imagination. If AI is framed mainly as a weapon to be won, countries may treat cooperation as weakness, smaller powers may be pushed into choosing sides, and safety work may become harder to coordinate.
From Cooperation To Rivalry
Between 2016 and 2020, Harding briefed politicians across the globe, including Barack Obama and Emmanuel Macron, on advances in AI. At Google DeepMind, her role involved thinking through ethical questions and possible risks around the technology.
Harding says the earlier approach to AI research “was rooted in international cooperation.” The focus, as she describes it, was on helping political leaders understand what AI might become, while recognizing that some risks would be better handled through collaboration across borders.
That atmosphere changed. The industry began to be described through contests between individual labs such as Anthropic and OpenAI, and through a wider rivalry between the US and China. The arms race metaphor became a common shorthand for explaining what was happening.
For Harding, that shorthand is not neutral. It turns a complex technology into a contest with winners and losers. Once policymakers accept that frame, they may become less willing to discuss shared standards, common safety goals, or broad access to benefits.
Why The War Metaphor Took Hold
Harding sees several forces behind the shift. One was a genuine belief that AI could be dangerous, especially in the wrong hands, and that democracies should therefore control the most powerful systems.
Another was political and commercial. Harding describes an anti-regulation current that benefited from portraying China as the central threat: “If you regulate us, you let China win.” In that framing, domestic oversight becomes a strategic liability rather than a normal part of governing powerful technology.
ChatGPT launched in November 2022 and brought far wider attention to AI. Harding also points to the wider moment in which that attention arrived: a global pandemic, growing anxiety about borders, and the war in Ukraine, where debates about AI, geopolitics, and weaponry became more concrete.
Those conditions made the arms race comparison easy to understand. AI was mapped onto the Cold War and discussed as akin to a nuclear weapon. Harding’s point is that a familiar historical analogy can feel clarifying while still producing a distorted view of the present.
Policy Follows The Frame
Harding argues that technology does not simply appear and change society on its own. Society also shapes technology. In the case of AI, she sees political culture in the US and the broader geopolitical environment as major forces influencing both AI policy and AI development.
She points to the Trump administration’s nationalist AI rhetoric and its effort to impose export controls on homegrown models as signs of the arms race frame becoming policy. In the source interview, WIRED also refers to Donald Trump issuing an executive order on AI in the last two weeks and his administration effectively forcing Anthropic to withdraw its latest frontier model from the market.
For European powers, concern about dependence on American technology is understandable in that context. Harding does not reject the importance of sovereign capacity in Europe and the UK. But she argues that isolationism has become too dominant as a policy driver.
The practical problem is that even the US and China cannot develop everything alone. Harding points to strategic chokepoints around chips, critical minerals, scientists, and markets. In her view, it is unrealistic to suggest that every country can build a completely sovereign AI stack.
A Role For Middle Powers
Harding’s answer is not to replace competition with idealistic cooperation. She argues that both can exist at the same time: “Competition is normal and healthy but doesn’t have to be mutually exclusive with collaboration and cooperation.”
One idea she has been calling for is a middle powers coalition. She gives a possible example involving Canada, France, Japan, South Korea, India, and the UK. The logic is leverage and scale: India has scale and diffusion of technology, the UK has talent and an advanced startup ecosystem, and Canada has critical minerals.
The deeper purpose is to resist the idea that AI is only a binary contest between two superpowers. If middle powers accept that story, Harding argues, they risk becoming smaller pieces on someone else’s board rather than actors with their own interests and bargaining power.
This matters especially for countries that import the technology. If the arms race frame dominates, they may feel pressure to align behind one superpower or another, even when that choice does not match their own interests.
The Risk Of A Narrow Future
Harding also says major labs are complicit in moving the rhetoric from collaboration toward competition. In her view, talking about AI as an arms race increases their power because it suggests the technology is so powerful, new, and unique that only they can manage it.
She does not say safe and responsible AI development is impossible under fierce competition. But she questions whether race language helps anyone plan calmly or collegially.
The destination she warns about is a more centralized future: excessive government control over AI systems, less safe and beneficial technology, and many smaller states forced to choose between superpowers. The lost opportunity, in her argument, is cooperation on areas such as security, food security, and ending disease.
That is why Harding’s critique is ultimately about more than wording. The phrase AI arms race may sound like a description of reality. Her warning is that, if enough leaders believe it, the metaphor can help make that reality true.