AI overlords say they want to democratise their technology. Here’s how | Richard Wilson and Clara Long

A fight over a datacentre in Utah has been a testing ground for citizens’ assemblies, paving the way to oversight

What’s your “(p)doom”, the percentage chance that rogue AI will create an existential catastrophe? Judging by their recent warnings, it is rising among tech bosses.

On 25 September, OpenAI paused training of its latest frontier models as reports of AI agents going rogue mounted. A day later, Australia’s Senate asked the bosses of OpenAI and Anthropic to appear before an inquiry after rogue OpenAI agents broke into US government websites and an Australian health portal. Both declined.

Bill Gates, meanwhile, has warned that AI in the wrong hands could “cause a billion deaths”. “No one thinks self-regulation is enough,” he said.

Days before, Dario Amodei and Sam Altman, the chief executives of Anthropic and OpenAI, appeared before the UN security council . Amodei warned that AI could become “a risk to humanity as a whole”. Altman said that “if AI is to be democratic, the most important decisions cannot be made by labs in San Francisco alone”.

Doomsday scenarios aside, AI is already disrupting schools, workplaces and public services, with the most marginalised often feeling the impacts harder and faster. If democracy is to matter in this era of tech power, surely we must try to democratise AI. But how?

Utah offers a glimpse. Proposals for a datacentre more than twice the size of Manhattan provoked a furious backlash over energy use and the potential impact on the state’s stressed water supplies and Great Salt Lake ecosystem, all while people felt excluded from having a say.

On 18 and 19 September, Utah Common Ground convened 38 residents selected by civic lottery to consider what role the public should have in shaping decisions about AI technology and infrastructure.

Despite their political differences, participants overwhelmingly agreed on the need for greater public control. Their 10 recommendations included public hearings and independent environmental assessments for hyperscale datacentres, community-led AI literacy programmes and tighter transparency rules. Every recommendation received at least 82% support. In a polarised state, that is a mandate legislators should struggle to wave away.

Globally, there have been more than 80 public participation projects focused on AI over the past five years, involving more than 200,000 people in more than 95 countries. In Switzerland, a citizens’ assembly demanded privacy, transparency and meaningful human control. A Belgian citizens’ panel called for AI that serves everyone and leaves nobody behind.

At present, however, this work is fragmented and largely ignored by decision-makers in government and the labs. The Citizens’ Track on AI , a decentralised network of assemblies rooted in communities but connected to the institutions where consequential decisions about AI are being made, intends to change that.

The Bloom Project , which helped convene the Utah assembly, is building one part of that network: a series of US local and state assemblies intended to feed into a national citizens’ assembly on AI in 2028. At the global level, the ambition is to connect assemblies across countries into the UN’s emerging processes for AI governance, while also creating direct lines of accountability to the AI companies.

A bridge built by the powerful can be withdrawn by them. That is why Utah matters. Its forum was rooted in civic groups that will outlast the forum itself, though it has no guaranteed hearing yet. Durable civic power needs both: communities that keep convening and decision-makers who have to answer them.

The same principle underpins proposals published last week for a standing global citizens’ assembly on AI : combining the independent moral authority of globally representative, self-governing citizens with legislated powers to hold AI companies accountable for their safety and alignment practices.

Can any of this protect us against AI being used to launch a cyber-attack, develop a biological weapon or escape human control altogether? Not by itself. But how much risk society should accept isn’t an engineering question, it’s a political one. Citizens’ assemblies can give governments the mandate, and the cover, to impose the testing, red lines and stop buttons that companies won’t impose on themselves.

The danger is not only that AI may escape human control. It is whether the humans controlling it are a tiny group of elite executives, investors and politicians vulnerable to corporate capture.

Altman was right: decisions of this magnitude cannot remain inside laboratories in San Francisco. But moving them into closed meetings between powerful states would not make AI democratic either.

Imagine instead every school, workplace and community connecting into this process: shaping what the AI companies do, but also becoming proactive in the face of the AI onslaught. A truly global citizens’ assembly rooted in the communities that understand the impacts of datacentres and labour displacement. Something large enough to counter the AI arms-race narrative that dominates debate and paralyses political action.

The key question, then, is not whether democratising AI is realistic, but whether AI can realistically be governed without it.

Democracy is not an optional restraint on this technology. It may be our only means of keeping its development and deployment under meaningful human control.

Richard Wilson is the CEO of Iswe Foundation and Clara Long is the co-executive director of the Bloom Project