Former DeepMind Executive: US Nationalist AI Policy Is Heading Toward Catastrophe

Verity Harding, a former DeepMind director, told Wired in an interview that the American government's nationalist course on artificial intelligence is no longer a theoretical danger but direct evidence that the worst-case scenario is starting to come true. She believes an AI arms race can end in catastrophe.

What Harding sees from the inside

Harding worked at DeepMind, Google's flagship AI lab and one of the most influential research organizations in the world. Over her years at the intersection of technology and policy, she developed a clear position: the main threat comes not from the technology itself but from how states build policy around it.

When governments view AI primarily through the lens of national superiority, the familiar logic of an arms race kicks in. "Beating the rival" becomes more important than "developing responsibly." Speed of development crowds out reflection on consequences. Secrecy beats transparency.

"The American government's nationalist attitude toward AI is evidence that the worst-case scenario is beginning to materialize,"

Harding told Wired.

Why AI nationalism is more dangerous than earlier arms races

Geopolitical rivalry in AI carries risks that differ fundamentally from earlier technology races. Nuclear technology was built behind high barriers and stayed in the hands of states. AI capabilities spread differently: through open publications and commercial APIs available to millions of developers worldwide.

AI nationalism creates several systemic problems at once:

• International exchange of information about risks and vulnerabilities is blocked for competitive reasons
• The pace of development is artificially accelerated, outrunning researchers' ability to assess the consequences
• Safety researchers lose influence over priorities that are increasingly dictated by geopolitics
• Global control mechanisms fail to take shape in time, because each side expects to be the one that sets the rules

States that treat AI as a geopolitical weapon objectively have no interest in safety measures that could slow their progress. According to Harding, it is this structural trap that breeds a potential catastrophe.

Technology versus geopolitics

Harding's position reflects a growing split in the AI community. Some experts see powerful models themselves as the main threat, given their capabilities and unpredictability. Others, like Harding, are convinced that the technology is neutral in itself; the problem is how political and military logic shapes its development.

When leading technological powers build national AI strategies on zero-sum logic, international safety forums turn into yet another arena of geopolitical competition rather than platforms for cooperation.

What this means

Harding's warning is part of a growing consensus among AI experts: as long as "AI nationalism" dominates international safety cooperation, the risk of catastrophic failure remains open. The technology is advancing faster than the mechanisms to control it, and that gap is what causes the greatest concern.