TL;DR

Ed Davey will close the Liberal Democrat conference in Brighton on Tuesday by urging world leaders to halt work on super-intelligent AI through a treaty modelled on nuclear non-proliferation. He will propose an international AI safety agency, headquartered in the UK and built along the lines of the IAEA, the UN’s nuclear watchdog. He will also say the government’s new US partnership leaves Britain leaning on a White House he casts as beholden to “tech bros”.

What Davey will propose

According to the Guardian’s preview, Davey will cast the race towards superintelligence as the modern equivalent of the race for nuclear weapons, and argue that the IAEA shows what governments can do together to contain a technology that could “destroy us all”. His case is that decisions about AI’s future currently rest with a handful of wealthy people in California and Beijing, with nothing to slow development down.

“We didn’t leave the atom bomb to a few billionaires to police themselves,” he is expected to say, naming Mark Zuckerberg, Sam Altman and Elon Musk as people the public cannot rely on.

He will target Andy Burnham’s new AI and autonomy partnership with Washington directly, saying it “doesn’t come close to meeting the scale of this threat”.

Where it sits in the wider debate

Davey is not alone in reaching for the IAEA model this week. At the UN in New York, Finland’s president Alexander Stubb, a lead organiser of a 22-nation declaration on keeping AI under human control, told POLITICO he would personally prefer a new body of that kind within the UN system. The UK has not yet signed the declaration.

The domestic backdrop matters too. Plans to require pre-release testing of frontier models fell away before Burnham took office, and the King last week warned AI chiefs of the technology’s “existential dangers”. A halt on superintelligence goes further than either.

Looking forward

Opposition parties do not make treaties, so the speech is best read as a position the Lib Dems will press in Parliament rather than a near-term policy. It does, however, make AI safety a point of party difference. Earlier in the conference, Max Wilkinson, the party’s home affairs lead, set out a levy on large tech firms and an agency for online crime including deepfake pornography, and pledged rules for the Meta smart glasses he labelled “perv goggles”. Whether the government responds on the international agency question, particularly with the UK’s G20 presidency approaching, is the thing to watch.