Verity Harding, Google DeepMind’s former head of global public policy, is arguing that the loudest metaphor in AI policy is doing real damage: the idea that artificial intelligence is an arms race.
Harding, who held the DeepMind policy role from 2016 to 2020, told WIRED that her work then involved explaining AI advances and risks to political leaders including Barack Obama and Emmanuel Macron. She said the field’s policy conversation was once more grounded in international cooperation. That has shifted toward rivalry, she argues, both between companies such as Anthropic and OpenAI and between the United States and China.
Harding has curated a new essay collection, Reframing the AI Arms Race, that makes the case that language is not decorative in this fight. The anthology includes contributions from historian Lawrence Freedman and Japanese politician Taro Kono, according to WIRED. Its shared premise is that describing AI like a weapon shapes what governments think they are allowed to do.
The policy problem with war metaphors
Harding’s warning is blunt enough: once governments accept the arms-race frame, cooperation starts to look like weakness. Safety testing, shared security work, and discussions about how benefits get distributed become harder to justify when every model release is treated as another missile in a silo.
She traces the turn toward that framing to several overlapping events. OpenAI released ChatGPT in November 2022, pulling public attention toward AI. Around the same period, the pandemic had already pushed governments back toward borders and national supply concerns, while Russia’s war in Ukraine made military uses of technology feel less abstract. Harding told WIRED that AI was then increasingly compared with Cold War nuclear competition.
She also described two political incentives behind the shift. Some officials and technologists genuinely believe powerful AI systems could be dangerous in hostile hands, leading them to argue that democracies should control the technology. Another camp, in Harding’s account, uses China as a convenient reason to resist regulation: restrict domestic AI companies, the argument goes, and Beijing benefits.
Smaller countries get squeezed
Harding does not reject national capacity. She said sovereign AI capability in Europe and the UK matters. Her objection is to isolationism becoming the whole policy program, since no country, including the US or China, controls every part of the stack.
The stack is not just models. It includes chips, critical minerals, researchers, customers, cloud infrastructure, and deployment channels. Harding’s point is that every country pretending it can build a sealed, domestic version of all of that is fantasy with a procurement budget.
She cited strategic chokepoints as the predictable result: one country limits access to chips, another can respond with minerals, talent, or markets. For countries outside the US-China rivalry, accepting the arms-race story can mean becoming a subordinate player in somebody else’s contest.
Harding told WIRED she has been calling for a coalition of “middle powers” such as Canada, France, Japan, South Korea, India, and the UK. In her framing, such a group would combine different strengths, including India’s scale, the UK’s talent and startup base, and Canada’s minerals.
She also accused major AI labs of benefiting from the race narrative because it concentrates authority around them. If AI is portrayed as uniquely powerful and dangerous, the companies building frontier systems can argue that only they understand the problem well enough to manage it.
Harding’s worst-case outcome is more government control, more centralization, and less useful safety cooperation on areas such as security, food security, and disease. That is a policy failure dressed up as strategic discipline, which is usually how bad metaphors get promoted.
This story draws on original reporting from WIRED.
