Urgent research needed to tackle AI threats, says Google AI boss

The world's most consequential technology debate now has a clear dividing line: America on one side, everyone else on the other. At this week's Delhi AI Impact Summit, the US delivered an unambiguous message through Michael Kratsios, head of the American delegation and White House technology adviser: "We totally reject global governance of AI." The statement marks a decisive break from collaborative efforts and raises profound questions about how the world's most powerful technology will be developed and deployed.

The timing couldn't be more striking. Whilst Kratsios insisted that "AI adoption cannot lead to a brighter future if it is subject to bureaucracies and centralised control", the summit heard repeated warnings from the very people building this technology that urgent oversight is essential. This creates an unusual dynamic where American technology leaders request oversight whilst their government explicitly rejects it.

When tech chiefs want what governments refuse

Speaking to the BBC in Delhi, Google DeepMind chief Sir Demis Hassabis called for more research on AI threats to be conducted "urgently" and advocated for "smart regulation" addressing "real risks". Sam Altman, the chief executive of OpenAI, echoed this view in his summit address, also calling for "urgent regulation". One reading is that this represents genuine concern from executives who understand the systems they're building better than anyone.

Another interpretation is less flattering: established players calling for regulation that could entrench their advantages and raise barriers for competitors. The timing of these calls is revealing. Both DeepMind and OpenAI are racing to deploy increasingly capable models whilst simultaneously warning about the need for guardrails.

Hassabis acknowledged to the BBC that keeping pace with AI development was "the hard thing" for regulators, a problem his company's rapid release schedule doesn't exactly help solve.

According to Hassabis, the technology poses two primary threats: misuse by bad actors and the risk of losing control as systems become more autonomous and powerful. He told the BBC his firm had a role in addressing these challenges but was "only one player in the ecosystem". When pressed on whether he could slow development to give safety researchers more time, he effectively demurred.

The China question driving everything

Behind America's regulatory stance sits a calculation about Beijing. Hassabis told the BBC that the US and the West maintain only a "slight" lead over China in AI capabilities, adding that Chinese researchers could catch up "in a matter of months". That assessment, whether accurate or not, provides the clearest explanation for Washington's position.

If the technological gap is measured in months rather than years, accepting international governance structures that could slow development becomes harder to justify through a competitive lens. Why agree to speed limits if your rival might ignore them? The US appears to be betting that maintaining a light-touch regulatory environment will help American companies move faster than Chinese counterparts operating under Beijing's heavier hand.

Whether this proves correct depends on assumptions about how much regulation actually slows innovation versus how much it prevents costly mistakes and public backlash that could trigger much harsher restrictions later. European nations and India, meanwhile, are pushing for coordinated frameworks. UK Deputy Prime Minister David Lammy told the summit that "security and safety must come first" and insisted politicians must work "hand in hand" with technology firms rather than leaving safety decisions to companies alone.

What the fracture means for business

For companies operating globally, divergent regulatory approaches create immediate complications. A technology deemed acceptable in the US might face restrictions in the EU or UK. Products designed for American rules might require expensive modifications for other markets.

Startups must decide whether to optimise for speed in permissive jurisdictions or compliance in stricter ones. The split also raises questions about research coordination. Safety research benefits from international collaboration and shared data. If American entities operate under fundamentally different assumptions about governance than their counterparts elsewhere, that cooperation becomes harder to sustain.

Investors face their own calculus. Regulatory arbitrage could offer American AI companies a near-term edge, making them attractive investments. But regulatory fragmentation also increases risk, as products might face sudden bans or requirements in major markets. The European Union's AI Act, which began enforcement this year, already demonstrates how different jurisdictions are establishing conflicting requirements.

Hassabis claimed to the BBC that his company doesn't "always get things right" but gets it "more correct than most" in balancing boldness with responsibility. By what measure remains unclear, and compared to whom.

The Delhi summit is expected to conclude with some form of joint statement, though its substance will likely reflect the lowest common denominator given America's position. The question facing other nations is whether to proceed with coordination amongst willing parties or abandon the effort entirely.

The US decision to reject global governance structures doesn't prevent other countries from establishing their own frameworks or collaborating amongst themselves. But it does guarantee that the world's largest technology companies, most of which are American, will face a fragmented landscape where the rulebook changes depending on geography. Whether that accelerates innovation or simply accelerates risk now depends on which government's bet proves correct.