How the global effort to keep AI safe went off the rails 

NEW DELHI — It started as an elite political conversation in an English country house to rein in this generation’s most powerful technology.

Now it’s a business dealmaking free-for-all in an Indian megacity. 

The annual global artificial intelligence summit, which takes place in New Delhi this week, has grown from 150 to 35,000 delegates in less than three years. In the process, the original motivation for the summit — as a global conversation to agree on safeguards to keep AI technology in check — has been relegated to quiet corners of the gathering.

On the surface, the Indian organizers of this year’s event have some aspects of AI safety in mind, at least according to the lofty slogans on street billboards dotted throughout the city.

Yet the final summit declaration set to be agreed later this week will fail to include the word “safety,” according to a draft text reported exclusively by POLITICO. After the first two summits addressed catastrophic risks to humanity and threats like job losses and environmental damage, India has instead switched the focus to practical applications and dealmaking.

With a White House friendly to the tech bros, a better understanding of how the technology will reshape global economies and a more complicated geopolitical environment, the conversations in New Delhi exemplify a stark shift: The global elite are no longer obsessing about how to control the risks of AI but figuring out who can benefit. 

“The summit reflects how the global AI conversation has shifted — from a narrow focus on safety to a broader push for impact,” said Jakob Mökander, director of tech policy at the Tony Blair Institute for Global Change, a think tank that has shaped the U.K. government’s AI policy. “Greater inclusion is a strength, but it inevitably makes focus harder.”

For those on the ground in the Indian capital, the gathering will be a marketplace for anyone to sell their version of what an AI-powered future should look like without having to address complex questions over how to ensure the rapid technological advances in AI land in a safe way.

Amber Sinha, executive director of digital rights group EDRi, described safety as “very much on the back burner,” a shift he said was already coming into view at the last summit in Paris. Sinha described the shift toward investments as “what we saw play out in Paris last year, and what, in even more stark terms, we’ve been seeing playing out now.”

Other participants were less pessimistic.

“You can take it the negative way, which is to say: The trust and safety conversation and the open-source conversation is in corners, and it’s sort of marginal and on the edges,” said Mark Surman, president of the Mozilla Foundation, which advocates an open internet and trustworthy AI. “On the other hand, I think what’s happened is … it’s just so big … that the conversation is broken into different pieces.”

While top entrepreneurs including OpenAI CEO Sam Altman, Google DeepMind CEO Demis Hassabis and Mistral CEO Arthur Mensch are expected on stage, political buy-in is mixed. France’s Emmanuel Macron is flying in with a high-level delegation, but countries including the U.K., the U.S. and Germany have a lower-level presence.

Mark Surman, president of the Mozilla Foundation , who says the AI safety conversation is “just so big … that the conversation is broken into different pieces.” | Dia Dipasupil/Getty Images

Every country for itself

At last year’s meeting in Paris, the U.K. — the summit’s original organizers — refused to sign the final declaration, in part because of how far it had strayed from the aims on safety. This year the U.K. has got on board with the trade-fair vibes. 

A delegation led by U.K. AI minister Kanishka Narayan and Deputy Prime Minister David Lammy will try to sell Britain as a destination for talent and investment. “The business leaders joining us in India will build concrete partnerships and secure investment that delivers opportunity for working people in the U.K., India and across the globe,” Lammy said.

The U.S., meanwhile, has arrived in New Delhi to sell its AI stack to the world through its AI Exports Program, the administration’s official policy to “extend American leadership in AI and decrease international dependence on AI technologies” — a reflection of its view of AI as a competition between its firms and China’s. 

The Trump administration “is very focused on encouraging the success of AI around the world. So this is an important opportunity and a strong showing for the U.S. government,” Undersecretary of Commerce for International Trade William Kimmitt said at an event in Washington this month.

American companies, including OpenAI, Anthropic and Meta, are announcing billions of dollars in investments in India this week — stealing a march on the Chinese, as Beijing is only sending a small delegation because the event clashes with Chinese New Year. 

The U.S. delegation is led by White House tech policy director Michael Kratsios, whereas Vice President JD Vance headed the last summit in Paris. On Thursday — the day world leaders including Indian PM Narendra Modi, Macron and Brazilian President Luiz Inácio Lula da Silva are meeting in New Delhi — President Donald Trump is in Washington to hold the inaugural meeting of his Board of Peace. 

That clash has already drawn former U.K. Prime Minister Tony Blair away from India, where he had been due to speak. His think tank, the Tony Blair Institute, now bills former U.K. Prime Minister Rishi Sunak (an adviser to Microsoft and Anthropic) and former U.K. Finance Minister George Osborne (managing director at OpenAI) on its lineup.

The EU, meanwhile, wants both to flex its position as a global regulator and to show it’s open to attract investment. That tension has been on display in recent months as EU authorities implemented the bloc’s flagship AI law while simultaneously rolling back safety provisions through a “simplification package,” after complaints from European companies that the law was too burdensome. 

EU tech chief Henna Virkkunen, the Commission’s representative at the summit, will “emphasize the EU’s ambition to accelerate AI deployment, scale innovation and work with trusted partners such as India to ensure AI remains human-centric, secure and aligned with democratic values,” the Commission’s tech department said ahead of the event.

Some argue the EU could gain traction at the summit for its governance model under which companies conduct voluntary assessments, including with the hosts. The regulatory approach of the Modi government has been “light-touch,” said the EDRi’s Sinha, with India in the lead-up to the summit announcing AI governance guidelines that focus on companies self-regulating.

Indian policymakers could take a page out of the Brussels playbook for its code of practice — the voluntary guidelines advising how companies should file risk assessments — “but I don’t see a lot of convergence besides from that,” Sinha said.

From Delhi to Davos?

As the week kicked off, tens of thousands of attendees spent at least part of their Mondays figuring out the Bharat Mandapam convention center — built for India’s 2023 G20 summit — and the adjacent 70,000 square meters in exhibition space. Some attendees faced long queues to enter the venue, while others flagged traffic issues and airport disruptions.

EU tech chief Henna Virkkunen, who is expected to “emphasize the EU’s ambition to accelerate AI deployment, scale innovation and work with trusted partners.” | Malin Wunderlich/picture alliance via Getty Images

Alongside the confusing logistics, questions also extended to the future of the summit series itself, given how far it has moved from its origins.

Some argue safety conversations may begin to take place elsewhere. “The summit series evolving to more of a global trade fair will push decisions on global AI governance into other forums,” said Mökander of the Tony Blair Institute. 

But Mozilla’s Surman argues the scale of the event in fact allows alternative tech players, coming from the “middle powers” referenced in Canadian PM Mark Carney’s anti-Trump speech at Davos, to find one another. That could give people choices beyond the mainstream tools in a field dominated by the biggest tech players from America.

With next year’s summit likely heading to Switzerland, the Davos World Economic Forum — a political and business whirlwind charged with setting the direction of politics and yet widely disparaged as a talking shop for global elites — is a relevant comparison. Another potential model is the United Nations’ decades-old series of global climate summits, known as COPs, which migrate from country to country each year.

“We need to choose whether we want to move toward a COP model, with the gradual formation of a body of texts, or toward a G7 model, with one country setting its priorities each year,” said Martin Tisné, a former French government envoy to the Paris AI summit. “Personally I think institutionalization [as with COP] is a very good idea to address the fragmentation of AI governance at the global level.”

“By passing the baton from one year to the next, supporting each other and contributing to coherence rather than fragmentation in global AI governance and infrastructure, an alliance is emerging, and it would be excellent news if Switzerland were to take up the torch in 2027,” Tisné said.

Pieter Haeck reported from New Delhi, Tom Bristow reported from London, and Océane Herrero reported from Mumbai. Katherine Long contributed reporting from Washington.