This week's big AI bosses call for slowdown, but disagree on the way

This week has brought unusual unanimity among the AI industry's most powerful voices: slow down. Anthropic CEO Dario Amodei kicked off with a call for a global slowdown because he fears software that can improve itself out of control. Within a few hours, OpenAI, Google DeepMind, Microsoft and Elon Musk all backed it, and later in the week Microsoft's Satya Nadella also joined the chorus.

Disagreement on the common course

But behind the shared headline lies disagreement about what a slowdown should actually mean. Sam Altman still promises rapid progress, even though he supports more caution and has introduced safety checks before major training runs. Elon Musk and Demis Hassabis only partially back it, and there are still no concrete measures. Mark Zuckerberg flatly refuses to put the whole industry on pause and believes each company is responsible for safety itself, while Meta will soon launch a new model. And from China comes the message that the slowdown proposal is merely a "Cold War playbook" aimed at the country, which shows that the AI race is at least as much about rhetoric and power as about real safety.

Politicians get involved

Politicians are getting involved too. Bernie Sanders and Steve Bannon, who otherwise rarely agree on anything, have jointly demanded tougher brakes in Washington, and OpenAI for the first time supports a bill on mandatory external safety checks. Sanders goes further and calls on Trump and Xi to negotiate a treaty that completely bans superintelligence, but so far that is only a political proposal. European Commission President Ursula von der Leyen warns that AI agents that "run away" from their boundaries are only a foretaste of what is coming and wants to use the EU's AI Act to set global standards.

At the same time, the three AI giants OpenAI, Anthropic and Google DeepMind are trying to preempt lawmakers by setting up a joint body themselves to test models before launch. Both Danish and international experts are skeptical of this kind of self-regulation.

The risk is not theoretical

That the risk is not just theoretical was underlined by a case from Spain, where an AI agent on its own found a security flaw, logged in and changed personal data without a human having asked it to. Here at home, the government is trying again with a bill that would give authorities a clear right to use AI with citizens' data, but the Danish Data Ethics Council still believes oversight is lagging.

Concrete rules are also appearing elsewhere: California has signed 13 new laws on children's online safety, including requirements for crisis protocols in AI chatbots, while New York is banning chatbots for the youngest schoolchildren.

Keep an eye on whether this week's fine words about slowing down are followed by concrete agreements, or whether it ends up as yet another chapter in the race between the great powers.