Verity Harding, a former head of global public policy at Google DeepMind, warns that framing artificial intelligence as an arms race is actively making the technology less safe.
The shift from cooperation to rivalry
Between 2016 and 2020, Harding briefed leaders including Barack Obama and Emmanuel Macron on the ethical risks of AI. She told WIRED that research was once rooted in international cooperation. That environment changed. Rivalries now dominate, pitting labs like Anthropic and OpenAI against each other, and the United States against China.
In her new anthology, Reframing the AI Arms Race, Harding and contributors such as historian Lawrence Freedman and Japanese politician Taro Kono argue that language drives policy. Describing AI as a weapon closes the door on the collaboration needed to manage risk and distribute benefits. Smaller nations following this framing often align with a superpower against their own interests.
Harding views the nationalist rhetoric of the Trump administration and its export controls on homegrown models as symptoms of this mindset. She believes a worst-case scenario is forming.
Where the war metaphor comes from
Harding told WIRED in early June that the war metaphor feels clarifying but restricts thinking. At DeepMind, her goal was to help leaders understand the technology’s capabilities and concerns. She observed a shift toward viewing the issue as a civilizational battle between the West and China.
Two forces drove this change. One was a belief that the technology was dangerous and democracies must control it. The other was an anti-regulation stream that pointed to China as a threat. The argument was that regulating the US would allow China to win.
ChatGPT, launched in November 2022, drew attention to AI. This coincided with the global pandemic and the war in Ukraine. These events made the discussion of AI weaponry feel very real. It quickly became accepted wisdom that AI was the new arms race, mapped onto the Cold War and compared to nuclear weapons.
Politics shaping the technology
Harding argues that the trajectory of any technology depends heavily on who leads global superpowers at pivotal moments. The political culture in the US is having a huge impact on AI policy and the resulting technology. The geopolitical environment is tense, partly due to an isolationist approach.
She advocates for an internationalist approach rather than countries turning inward. Sovereign capacity in Europe and the UK remains important. However, isolationism has become the only driving force in AI policymaking, obscuring other possibilities.
Even the US and China cannot develop everything themselves. This creates strategic chokepoints. One nation might deny access to chips, critical minerals, or scientists. It is unrealistic to suggest every country can maintain a completely sovereign AI stack.
In the last two weeks, Donald Trump issued an executive order on AI dripping with nationalist rhetoric. His administration effectively forced Anthropic to withdraw its latest frontier model from the market.
A middle powers coalition
Harding does not see the internationalist approach as opposing the need for sovereignty. Competition can be normal and healthy without being mutually exclusive with collaboration.
She is calling for a middle powers coalition involving Canada, France, Japan, South Korea, India, and the UK. India offers scale and technology diffusion. The UK provides talent and an advanced startup ecosystem. Canada holds critical minerals.
The goal is to avoid the belief that AI is a binary race between two superpowers. Accepting that narrative makes it true. It turns smaller nations into lesser chess pieces on one side or the other.
Money certainly influenced the speed and freneticism of the industry rush. But funding alone does not account for the shift in rhetoric. Major labs are complicit. Talking about AI as an arms race accrues power to them by suggesting only they have the answer or should be in charge of the solution.
What happens next
It is possible to balance competition with safe development. But the language of a race does not help anyone plan calmly or collegially.
Continuing down this road leads to excessive government control and centralization of power. Systems become less safe and beneficial because collaboration on security, food security, or ending disease fails. Many nations become vassal states forced to pick one superpower.
Harding notes that even small cooperative tasks require effort. The arms race narrative and competition between labs mean people no longer want to exercise that muscle. Without use, the ability to cooperate will wither.
What it means
Practitioners and developers must stop treating AI policy as a zero-sum game. If nations accept the arms race framing, they lose the ability to negotiate security standards or share resources. The immediate practical step is for medium-sized economies to form their own coalitions to demand a seat at the table, rather than waiting for instructions from Washington or Beijing.
