How to make AI safe—and free of America and China
Leaders | Sovereign default
How to make AI safe—and free of America and China
State-backed efforts to catch up with frontier models are doomed
Share
Illustration: John Provencher
Jul 15th 2026|4 min read
PROGRESS IS rapidly being made towards regulating the release of powerful new AI models in America. Its government is set to soon outline its plan after taking suggestions from the industry. Speaking to The Economist this weekSir Demis Hassabis, the boss of Google DeepMind, pitched a hybrid public-private regulator for the industry, modelled on the agency that regulates brokers and stock markets. The idea is similar to our own thoughts.
Rules are needed because the newest frontier models possess some dangerous powers—from hacking critical digital infrastructure to writing recipes for new bioweapons. Without any organised way of dealing with such risks, the government has taken to making up the rules as it goes along in order to control access to Mythos and Sol, Anthropic and OpenAI’s latest offerings. Ideally, predictable rules would be agreed on internationally, especially between America and China, whose models are ahead of everyone else’s. But time is short. Sir Demis thinks that, if America moves unilaterally, the rest of the world will sign up to its system.
---
ShareReuse this content
Donald Trump’s blind alley
America’s president looks bereft of good options for solving the stand-off in the Gulf
Who is capable of evil?
Stop lowering the age of criminal responsibility
---
A no-brainer for protecting your brain
One simple vaccination may dramatically reduce the risk of dementia
---
England needs fewer council homes, not more
Andy Burnham’s plan is no way to ease the housing crisis
Two cheers for Trump Accounts
The grubby scheme contains the seeds of a good idea
The man who would change Russia
A leading oligarch speaks out, warning of the looming disaster facing his country
---
---